Vibration Analysis and Control PDF

VIBRATION ANALYSIS AND
CONTROL NEW TRENDS

AND DEVELOPMENTS

Edited by Francisco Beltrn-Carbajal

Vibration Analysis and Control New Trends and Developments

Published by InTech
Janeza Trdine 9, 51000 Rijeka, Croatia

Copyright 2011 InTech
All chapters are Open Access articles distributed under the Creative Commons
Non Commercial Share Alike Attribution 3.0 license, which permits to copy,
distribute, transmit, and adapt the work in any medium, so long as the original
work is properly cited. After this work has been published by InTech, authors
have the right to republish it, in whole or part, in any publication of which they
are the author, and to make other personal use of the work. Any republication,
referencing or personal use of the work must explicitly identify the original source.

Statements and opinions expressed in the chapters are these of the individual contributors
and not necessarily those of the editors or publisher. No responsibility is accepted
for the accuracy of information contained in the published articles. The publisher
assumes no responsibility for any damage or injury to persons or property arising out
of the use of any materials, instructions, methods or ideas contained in the book.

Publishing Process Manager Masa Vidovic
Technical Editor Teodora Smiljanic
Cover Designer Jan Hyrat
Image Copyright Igor Klimov, 2010. Used under license from Shutterstock.com

First published August, 2011
Printed in Croatia

A free online edition of this book is available at www.intechopen.com
Additional hard copies can be obtained from orders@intechweb.org

Vibration Analysis and Control New Trends and Developments,
p. cm.
ISBN 978-953-307-433-7

free online editions of InTech
Books and Journals can be found at
www.intechopen.com

Contents

Preface IX
Chapter 1 Adaptive Tuned Vibration Absorbers:
Design Principles, Concepts
and Physical Implementation 1
Philip Bonello
Chapter 2 Design of Active Vibration Absorbers
Using On-Line Estimation
of Parameters and Signals 27
Francisco Beltran-Carbajal, Gerardo Silva-Navarro,
Benjamin Vazquez-Gonzalez and Esteban Chavez-Conde
Chapter 3 Seismic Response Reduction of Eccentric
Structures Using Liquid Dampers 47
Linsheng Huo and Hongnan Li
Chapter 4 Active Control of Human-Induced Vibrations
Using a Proof-Mass Actuator 71
Ivn M. Daz
Chapter 5 Control Strategies
for Vehicle Suspension System Featuring
Magnetorheological (MR) Damper 97
Min-Sang Seong, Seung-Bok Choi and Kum-Gil Sung
Chapter 6 A Semiactive Vibration Control Design
for Suspension Systems with Mr Dampers
Hamid Reza Karimi 115
Chapter 7 Control of Nonlinear Active Vehicle Suspension
Systems Using Disturbance Observers 131
Francisco Beltran-Carbajal, Esteban Chavez-Conde,
Gerardo Silva Navarro, Benjamin Vazquez Gonzalez
and Antonio Favela Contreras
VI Contents

Chapter 8 Semi-Active Control of Civil Structures Based
on the Prediction of the Structural Response:
Integrated Design Approach 151
Kazuhiko Hiramoto, Taichi Matsuoka and Katsuaki Sunakoda
Chapter 9 Seismic Response Control Using Smart Materials 173
Sreekala R, Muthumani K, Nagesh R Iyer
Chapter 10 Whys and Wherefores of Transmissibility 197
N. M. M. Maia, A. P. V. Urgueira and R. A. B. Almeida
Chapter 11 Control Design Methodologies for Vibration
Mitigation on Wind Turbine Systems 217
Ragnar Eide and Hamid Reza Karimi
Chapter 12 Active Isolation and Damping of Vibrations for
High Precision Laser Cutting Machine 243
Andrea Tonoli, Angelo Bonfitto and Mario Silvagni
Chapter 13 Bearings Fault Detection Using Inference Tools 263
Miguel Delgado Prieto, Jordi Cusid i Roura and
Jose Luis Romeral Martnez
Chapter 14 Vibration Analysis of an Oil Production Platform
Submitted to Dynamic Actions Induced by
Mechanical Equipment 281
Jos Guilherme Santos da Silva, Ana Cristina Castro
Fontenla Sieira, Luciano Rodrigues Ornelas de Lima
and Bruno Dias Rimola
Chapter 15 MIMO Vibration Control for a Flexible Rail Car Body:
Design and Experimental Validation 309
Alexander Schirrer, Martin Kozek and Jrgen Schftner
Chapter 16 Changes in Brain Blood Flow on Frontal Cortex Depending
on Facial Vibrotactile Stimuli 337
Hisao Hiraba, Takako Sato, Satoshi Nishimura, Masaru Yamaoka,
Motoharu Inoue, Mitsuyasu Sato, Takatoshi Iida, Satoko Wada,
Tadao Fujiwara and Koichiro Ueda

Preface

This book focuses on the very important and diverse field of vibration analysis and
control. The sixteen chapters of the book written by selected experts from international
scientific community cover a wide range of interesting research topics related to original
and innovative design methodologies of passive, semi-active and active vibration control
schemes, dynamic vibration absorbers, vehicle suspension systems, structural vibration,
identification, vibration control devices, smart materials, fault detection, finite element
analysis and several other recent practical applications and theoretical studies of this
fascinating field of vibration analysis and control. The book is addressed not only to both
academic and industrial researchers and practitioners of this field, but also to
undergraduate and postgraduate engineering students and other experts and
newcomers in a variety of disciplines seeking to know more about the state of the art,
challenging open problems, innovative solution proposals and new trends and
developments in this area.
The book is organized into 16 chapters. A brief description of every chapter follows.
Chapter 1 presents the basic design principles of adaptive tuned vibration absorbers
(ATVA) and a comprehensive review of the various design concepts that have been
presented for the ATVA, including the latest innovations contributed by the author.
Chapter 2 introduces a design approach for active vibration absorption schemes in
linear mass-spring-damper mechanical systems subject to exogenous harmonic
vibrations, which can simultaneously be used for vibration attenuation and desired
reference trajectory tracking tasks. Chapter 3 deals with the seismic response control of
eccentric structures using Circular Tuned Liquid Column Dampers (CTLCD). The
optimal control parameters are derived from the motion equation of the CTLCD-
structure system and supposing that ground motion is a stochastic process. Chapter 4
presents the practical implementation of an active mass damper to cancel excessive
vertical vibrations on an in-service office floor and on an in-service footbridge. In
Chapter 5, the authors describe the formulation and experimental evaluation of
various vibration control strategies for semi-active vehicle suspension system with
magnetorheological (MR) dampers. The design of a back-stepping control scheme for
semi-active vehicle suspension systems with MR dampers is the focus of Chapter 6.
Chapter 7 proposes a robust control scheme based on the real-time estimation of
perturbation signals for active nonlinear or linear vehicle suspension systems subject
to unknown exogenous disturbances due to irregular road surfaces. Chapter 8
X Preface

introduces a methodology for semi-active control system design of civil structures. A
semi-active control law based on a one-step-ahead prediction of the seismic response
is proposed, which employs a vibration control device developed by the authors.
Chapter 9 deals with the seismic response control using smart materials. Mathematical
models are developed to predict the maximum energy dissipation capability of the
material under study. Chapter 10 presents a general overview on the transmissibility
concept for multiple degree-of-freedom systems. It is shown that the various ways in
which transmissibility can be defined and applied opens various possibilities for
research in different domains, like system identification, structural modification,
coupling analysis, damage detection, model updating, vibro-acoustic applications,
isolation and vibration attenuation. The focus of Chapter 11 is on design of control
schemes for vibration mitigation on wind turbine systems. The main control objective
of the proposed schemes is to reduce the torque variations by using speed control with
collective blade pitch adjustments. Chapter 12 focuses on the evaluation of an active
isolation and vibration damping device on the working cell of a micro-mechanical
laser center, using active electromagnetic actuators. Chapter 13 presents an overview of
multisensory inference approaches used to characterize motor ball bearings, and their
application to a set of motors with distributed fault failure. The results show that a
multivariable design contributes positively to damage monitoring of bearings. Chapter
14 investigates the dynamic behavior of an oil production platform submitted to impacts
produced by rotating machinery. A computational model is developed for the structural
system dynamic analysis. The peak acceleration values and maximum displacements
and velocities are employed to evaluate the structural model performance in terms of
human comfort, maximum tolerances of the mechanical equipment and vibration
serviceability limit states of the structural system. Chapter 15 proposes LQG and
weighted H2 MIMO control design methods for the vibration control of lightweight rail
car body structures. These designs are studied and compared to achieve vibration
reduction and passenger ride comfort improvement in a highly flexible metro rail car
body. The metro car body structure is directly actuated via locally mounted Piezo stack
actuators. Chapter 16 concludes the book, describing a study on vibrotactile stimuli on
the submandibular glands stimulated by vibration with one motor and two motors.
Finally, I would like to express my sincere gratitude to all the authors for their
excellent contributions, which I am sure will be valuable to the readers. I would also
like to thank the editorial staff at InTech for their great effort and support in the
process of edition and publication of the book.
I truly hope that this book can be useful and inspiring for contributing to the
development of technology, new academic and industrial research and many
inventions and innovations in the field of vibration analysis and control.

Francisco Beltrn Carbajal
Energy Department
Azcapotzalco Unit of the Autonomous Metropolitan University
Mexico

1
Adaptive Tuned Vibration Absorbers:
Design Principles, Concepts and
Physical Implementation
Philip Bonello
University of Manchester
United Kingdom
1. Introduction
The tuned vibration absorber (TVA) has been used for vibration control purposes in many
sectors of civil/automotive/aerospace engineering for many decades since its inception by
(Ormondroyd & Den Hartog, 1928). A tuned vibration absorber (TVA), in its most generic
form, is an auxiliary system whose parameters can be tuned to suppress the vibration of a
host structure. The auxiliary system is commonly a spring-mass-damper system (or
equivalent) and the TVA suppresses the vibration at its point of attachment to the host
structure through the application of an interface force. The tuned frequency
a
of the TVA
is defined as its undamped natural frequency with its base (point of attachment) blocked.
The TVA can be used in two distinct ways, resulting in different optimal tuning criteria and
design requirements (von Flotow et al., 1994):
a. It can be tuned to suppress (dampen) the modal contribution from a specific troublesome
natural frequency
s
of the host structure over a wide band of excitation frequencies.
b. It can be tuned to suppress (neutralise) the vibration at a specific troublesome excitation
frequency , in which case it acts like a notch filter.
When used for application (a), the TVA referred to as a tuned mass damper (TMD).
a
is
optimally tuned to a value slightly lower than that of the targeted mode
s
and an optimal
level of damping needs to be designed into the absorber. When used for application (b), the
TVA is referred to as a tuned vibration neutraliser (TVN) (Brennan, 1997, Kidner &
Brennan, 1999) or undamped TVA. The optimal tuning condition is in this case is
a
=
and the TVN suppresses the vibration over a very narrow bandwidth centred at the tuned
frequency. Total suppression of the vibration at this frequency is achieved when there is no
damping in the TVN.
Deviation from the tuned condition (mistuning) degrades the performance of either variant
of the TVA (von Flotow et al., 1994) and it can be shown that a mistuned vibration
neutraliser could actually increase the vibration of its host structure (Brennan, 1997). To
avoid mistuning, smart or adaptive tunable vibration absorbers (ATVAs) have been
developed. Such devices are capable of retuning themselves in real time. Adaptive
technology is especially important in the case of the TVN since the low damping
requirement in the spring element can raise the host structure vibration to dangerous levels


2
in the mistuned condition. In this case, mistuning can occur either due to a drift in the
forcing frequency or due to a drift in tuned frequency caused by environmental factors (e.g.
temperature change). Hence, a TVN needs to be adaptive to be practically useful.
At the heart of an ATVA is a means for adjusting the tuned frequency
a
in real time.
This is frequently done through the variation of the effective mechanical stiffness of the
ATVA, although other means are possible. Whatever the retuning method used, the
device should be tunable over an adequate range of frequencies, and the adjustment
should be rapid and with minimum power requirement. The device must also be cheap
and easy to manufacture. To maximise vibration attenuation, the retuning mechanism
should add as little as possible to the redundant mass of the device and, in the case of the
neutraliser, have a low structural damping (Brennan, 1997). The technical challenge is to
design an adaptive device with such attributes.
This chapter continues with a quantitative illustration of the basic design principles of
both variants of the TVA. It will then present a comprehensive review of the various
design concepts that have been presented for the ATVA, including the latest innovations
contributed by the author. This section will cover the use of piezoelectric actuators, shape-
memory alloys and servo-actuators within the smart structure of the ATVA. Control
algorithms and their implementation through MATLAB
with SIMULINK
will also be
discussed.
2. Basic design principles of TVA
With reference to Fig. 1a, the above-defined frequency
a
coincides exactly with the lowest
anti-resonance frequency of the attachment point receptance frequency response function
(FRF) of the undamped free-body TVA structure, ( )
AA A
r Y F = , where
A
Y and F are
complex amplitudes of
A
y and the interface force ( ) f t , for harmonic vibration at circular
frequency . It is for this reason that, for the TVN, the condition
a
= defines optimal
tuning. For excitation frequencies below the first non-zero resonance frequency
m
of
( )
AA
r , the absorber can be represented by the equivalent two-degree-of-freedom model
shown in Fig. 1b (Bonello & Groves, 2009). Fig. 1b shows that the absorber mass
a
m is split
into an effective mass
, a eff a
m Rm = and a redundant mass ( )
,
1
a red a
m R m = . The latter mass
simply adds to the host structure. The effective part of the absorber is the single-degree-of-
freedom system constituted by mass
, a eff
m , the spring of stiffness
2
, a a a eff
k m = and a
damping element. In the case of a TMD, where damping is deliberately designed into the
device, it is preferable to represent it by viscous damper of frequency-independent
coefficient
a
c . In the case of a TVN, where the damping is an unwanted inherent feature of
the spring element, it is best represented by a structural damping mechanism of loss factor
a
, for which the equivalent viscous damping coefficient is
a a
k . The method for
deriving the equivalent two-degree-freedom is detailed in Section 4.1.
2.1 Tuned mass damper
The purpose of the TMD is to dampen a particular resonance peak of the FRF ( )
AP
r
connecting the response at A to an external force ( )
p
f t applied to the host structure at some
arbitrary point P. It is useful for applications where the excitation has a broad frequency
spectrum containing the targeted mode. The damping in the original host structure (i.e. the

Adaptive Tuned Vibration Absorbers: Design Principles, Concepts and Physical Implementation

3
structure prior to the addition of the TMD) is commonly assumed to be negligible relative to
that introduced by the TMD. Suppose the TMD is targeted at the s
th
mode of the original
host structure. Let
s
be the frequency of this mode and
( ) s
A
denote the value of the
corresponding mass-normalised mode-shape at the degree of freedom being targeted (e.g.
the vertical displacement at A in Fig. 1). Then, from standard modal theory (Ewins, 1984),
the contribution of the targeted mode to the dynamics of the original host structure at the
targeted degree of freedom can be represented as the simple mass-spring system of mass
( ) ( )
{ }
2
1
s s
A A
M = and stiffness
( ) ( ) 2
s s
s A A
K M = (see Fig. 2(a)). The dynamics of the original host
structure at the targeted degree of freedom can be accurately modelled in this form for
excitation frequencies in the vicinity of
s
, where the targeted mode is dominant.
Addition of the TMD in Fig. 1b to the system in Fig. 2(a) results in the system in Fig. 2(b).
Notice that original host structure modal mass
( ) s
A
M needs to be readjusted to
( )
,
s
a red A
M m + ,
to account for the addition of the redundant absorber mass to the host structure. The
harmonic analysis of the systems in Figs. 2(a,b) (i.e. analysis with
{ }
j
Re e
t
P P
f F

= ) then
gives a modal approximation of ( )
AP
r , denoted by
( )
( )
s
AP
r , which is accurate for
frequencies in the vicinity of
s
.

Fig. 1. Generic TVA

Fig. 2. Dynamic modal model of the host structure without/with TMD for frequencies in
the vicinity of
s

( ) s
A
K

a
y

( )
red a
s
A
m M
,
+
( )
( )
( ) t f
P s
A
s
P

a
k

a
c

eff a
m
,

A
y

( ) s
A
M

( ) s
A
K

A
y

( )
( )
( ) t f
P s
A
s
P

(a) Host structure
(b) Host structure with TMD


4
( )
( )
s
AP
r can be dampened by adapting the harmonic analysis in (Den Hartog, 1956) to the
system in Fig. 2(b). The optimal tuning condition is found to be:

1
1
opt
a
s

=
+
(1)
where

( ) ( )
{ } ,
s s
s s a red A A
M M m = + ,
( )
{ } , ,
s
a eff a red A
m M m = + (2)
The viscous damping coefficient is given by:

,
2
a a a eff a
c m = (3)
The optimal value of the viscous damping ratio
a
is given by (Den Hartog, 1956):

( )
3
3
8 1
opt
a

=
+
(4)

The optimisation of the modulus of
( )
( )
s
AP
r is illustrated in Fig. 3. The introduction of the
TMD splits the original host structure resonance peak into two peaks separated by an anti-
resonance. Points M and N are referred to as the fixed points since, for given and
a
,
the function
( )
( )
s
AP
r of the modified structure passes through them regardless of the value
of
a
. The optimal tuning condition of eq. (1) ensures that the fixed points are level with
each other (this is the case illustrated in Fig. 3). With this condition in place, the optimal
damping condition of eq. (4) ensures that the peaks of
( )
( )
s
AP
r coincide as closely as
possible with the fixed points. The height of these optimised peaks is approximately
inversely proportional to .
As an illustration of the effect of a TMD on a multimodal (i.e. continuous) system, consider a
TMD attached to the tip of a cantilever and tuned to attenuate its second flexural mode (Fig.
4). The cantilever OA is of length 1m and made of steel (Young Modulus 200 GN/m
2
,
density 7850 kg/m
2
) of circular section with diameter 3cm.
Assuming an Euler-Bernoulli beam model, an eigenvalue analysis yields ( )
2
2 = 132.8
Hz,
( ) 2
1.387
A
M = kg. The mass ratio is taken as 2%. For simplicity, the TMD is assumed to
have no redundant mass. The TMD parameters were computed according to the above
formulae. The receptances ( )
AP A P
r Y F = of the system with and without the TMD were
evaluated using the Dynamic Stiffness Method (Bonello & Brennan, 2001) and shown in Fig.
5, where the dampening of the targeted resonance is clearly evident.
Fig. 6 illustrates the effect of mistuning i.e. deviation from the tuned condition of eq. (1). The
stiffness
a
k of the TMD was varied such that it was mistuned by 10% i.e.
a
was set to
1.1
opt
a
, with
a
c kept the same as the optimal case of Fig. 5 (considering eqs. (3) and (4), this
means that
a
of the mistuned case in Fig. 6 is not
opt
a
). It is clear from Fig. 6 that a slight
mistuning produces significant deterioration of the TMD performance.


5

0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2
0
10
20
30
40
50

Fig. 3. Effect of damping on the modal approximation of the attachment point FRF (case
shown is for the tuned condition, eq. (1), with 0.02 =

Fig. 4. TMD attached to a cantilever

0 50 100 150 200 250 300 350 400
10
-6
10
-4
10
-2

Fig. 5. Effect of TMD targeted against the second flexural mode of cantilever in Figure 4:
original system (dashed line); with TMD (solid line)
a
k
a
c
eff a
m
,
{ }
t
P p
F f
j
e Re =
O
A
P
A
y 0.4m 0.6m
( )
AP
r
(m/N)
( ) 2 (Hz)
0 =
a
( )
( )
s
AP
r

s

a
=
M N
No TMD
( )
( )
( )
s
A s
P
s
A
K
opt
a a
=

opt
a a
2 =

opt
a a
5 . 0 =


6
0 50 100 150 200 250 300 350 400
10
-6
10
-4
10
-2

Fig. 6. Effect of a mistuned TMD on cantilever in Figure 4: TMD optimally tuned and
damped as per eqs. (1) and (4) (solid line); 10% mistuned TMD (dotted line)
Finally, it is worth mentioning that a similar effect to the TMD can be achieved through an
electrical analogue, wherein the auxiliary system is a piezoelectric shunt circuit (Park, 2002).
In such an electrical TVA, a piezoelectric patch is bonded to the host structure and
connected across an external inductor-resistor circuit. The piezoelectric patch is used to
convert the vibration energy of the host structure into electrical energy and introduces a
capacitor effect into the circuit, turning it into an L-C-R circuit. The electrical energy is then
dissipated most efficiently as heat through the resistor when the electrical resonance
produced by the LC components is close to the frequency of the targeted mode and the
resistor has an optimum value. One major disadvantage of the electrical TVA is the
difficulty in deriving the transfer function of the modified system (on which the
optimisation is based); the difficulty increases with the complexity of the host structure. For
this reason the electrical TVA has typically been restricted to simple host structures like
cantilevers (e.g. Park, 2002). In contrast, the classical theory of the mechanical TMD is
readily applicable to any arbitrary host structure since the only host structure data it
requires are the frequency and modal mass of the targeted mode.
2.2 Tuned vibration neutraliser
The purpose of the TVN is to plant an anti-resonance in the FRF ( )
AP
r at some particular
chosen value of the excitation frequency . Hence, the TVN is typically used for
applications where the excitation is entirely, or mainly, at a single frequency (i.e. harmonic).
For example, suppose that it is required to cancel the tip vibration of the above considered
cantilever (Fig. 4) at a frequency of 50 Hz. The optimal condition for a TVN is

a
= (5)
Hence, in this example,
a
is optimally set to 100 . Neglecting the redundant TVA mass
and assuming an effective mass of the absorber, the absorber stiffness is calculated
accordingly. The effect of the TVN is illustrated in Fig. 7, where the absorber mass is
assumed to be 2% of the total mass of the beam. It is seen that an anti-resonance is
introduced at the desired frequency, in addition to a resonance at a slightly higher
( )
AP
r
(m/N)
( ) 2 (Hz)


7
frequency (52 Hz) (the difference between the anti-resonance and resonance frequencies is
found to increase with the effective TVN mass).
(Brennan, 1997) defines the attenuation D provided by a TVN for harmonic excitation as the
ratio of the vibration amplitude at A without the TVN to the amplitude there with the TVN
attached and optimally tuned:

free
TVN , opt
A
A
Y
D
Y
= (6)
In the absence of damping in the absorber D (complete attenuation). The attenuation
degrades with increasing absorber damping
a
(since this reduces the depth of the anti-
resonance in Fig. 7). Also, for given absorber damping, the TVAs attenuating capability
degrades as
, a eff
m is reduced. In fact, for a host structure that is a rigid machine of mass M
mounted on soft isolators, (Brennan , 1997) showed that

a
D (7)
where

( )
, , a eff a red
m M m = + (8)

Deviation from the optimally tuned condition
a
= (mistuning) can occur due to a change
in the excitation frequency (e.g. a change in operating speed of rotating machinery). It is
evident from Fig. 7 that even a slight mistuning will drastically degrade the performance of
the TVN. In fact, as can be seen in Fig. 7, if the excitation frequency drifts above
a
then the
vibration neutraliser actually increases the vibration of its host structure due to the extra
resonance it introduced into the system. This extra resonance is itself made more
pronounced by the low damping requirement.

0 50 100 150 200 250 300 350 400
10
-6
10
-4
10
-2

Fig. 7. Effect of TVN tuned to an excitation frequency of 50Hz on the cantilever in Fig. 4:
original system (dashed line); with TVN (solid line)
( ) 2 (Hz)
( )
AP
r
(m/N)
a
=


8
3. Adaptive tuned vibration absorbers an overview
Tuning a TVA involves making the appropriate adjustment of
a
and this is done
through an adjustment in one or more properties of the TVA structure. Mistuning is avoided
through the use of adaptive (or smart) tuneable vibration absorbers (ATVAs) which can
automatically perform the necessary adjustment in real time (von Flotow et al., 1994,
Brennan et al., 2004a). As demonstrated in the previous section, mistuning is a far more
serious issue for the TVN, since the requirement for low absorber damping can raise the
host structure vibration to dangerous levels in the mistuned condition. It is for this reason
that adaptive technology has been mainly developed in the context of the TVN. The ATVAs
considered in the remainder of the chapter will therefore exclusively be vibration neutralisers.
In the context of the TVN, adaptive tuning of the device involves maintaining the condition
a
= in the presence of variable conditions (typically a time-varying excitation frequency
, in which case the antiresonance in Fig. 7 is shifted in real time along the frequency axis,
in accordance to the current value of the excitation frequency). The challenge for ATVA
designers is to produce a device with the following attributes:
i. low structural damping;
ii. any actuating mechanism to retune the device should add as little as possible to the
redundant mass;
iii. the device should be tuneable over a wide range of frequencies;
iv. retuning should be rapid and with minimum power requirement;
v. the device should be cheap and easy to design and manufacture.
Various design concepts for ATVAs have been proposed (von Flotow et al., 1994, Brennan et
al., 2004a). One early variable stiffness element used in a vibration absorber was described
in (Longbottom et al., 1990). A mass was sandwiched between a pair of pneumatic rubber
bellows (Fig. 8) and the stiffness was adjusted by changing the air pressure inside the
bellows. Further work on this device by (Long et al., 1998) resulted in a means of
automatically adjusting the stiffness. However, the high amount of damping introduced by
the rubber bellows was a major disadvantage.

Fig. 8. Pneumatic rubber bellows ATVA (Brennan et al., 2004b)


9
One recent strategy for adaptation, by (Rustighi et al., 2005), was to utilise the variation with
temperature of the Young Modulus of a beam-like ATVA made of a shape memory alloy
(SMA) conductor (Fig. 9). The SMA wire formed a double cantilever, projecting from either
side of the central attachment point to the host structure. The ATVA stiffness was controlled
by adjusting the current through the wire. Despite being strong in attributes (i), (ii) and (v)
above, this device could only achieve a maximum variation of around 20% in tuned
frequency, taking as long as 2 minutes with a 9A current to do so (Brennan et al., 2004a,
Rustighi et al., 2005).

Fig. 9. Shape memory alloy ATVA (Rustighi et al., 2005)
Another recent approach, by (Bonello et al., 2005) was to utilise a mass-spring ATVA in
which the stiffness element was composed of parallel curved beams in longitudinal
compression (Fig. 10). The longitudinal stiffness was controlled by adjusting the curvature
through piezo-ceramic actuators bonded to the beams. This device was capable of very
rapid tuning over a frequency range 36-56 Hz (56% variation). However, this design concept
was inherently limited to low frequency applications as a result of inertia effects in the
curved beams (Bonello et al., 2005).
Other works have focused on the use of a beam-like ATVA controlled through servo-
actuation. This concept remains the best approach for applications requiring a wide tuning
frequency range (Carneal et al., 2004). Figs. 11-15 show various such ATVA designs. The
moveable-supports ATVA in Fig. 11 was patented by (Hong & Ryu, 1985). It consisted of a
beam with a mass attached to its centre and supports that could be moved relative to each
other, thereby altering
a
.
(Brennan, 2000) performed a theoretical study of the tuning frequency characteristics of the
designs in Figs. 12, 13. In Fig. 12, the ATVA beam is composed of two beams and the
effective stiffness of the ATVA is adjusted by pushing apart the two constituent beams at the
centre, thereby altering the ATVA beam cross-section. Such a device was built and tested in
(Kidner et al., 2002), where a maximum adjustment of 35% in tuned frequency was
achieved. This concept appears to provide the most rapid tuning of all beam-like designs in
Figs. 11-15 since the actuator is required to move the least distance to achieve a given change


10
in
a
. However, the actuator has to work against much larger forces and the variability in
a
is clearly limited by the maximum deformation that the constituent beams can
withstand as they are being prised apart. Through elementary analysis, (Brennan, 2000)
predicted that a considerably greater tuning range is achievable through the two alternative
designs in Fig. 13 (moveable beam or moveable masses). In the moveable beam
approach the masses are fixed relative to the beam and the beam lengthens or shortens. This
device does not appear to have been built and would require some form of telescoping
beam. The moveable-masses concept is more feasible: the beam is of fixed length and
tuning is achieved by repositioning the attached masses. (Von Flotow et al., 1994) describes
devices that appear to match this latter description in operation on the Boeing Chinook
helicopter, although the details available are very sketchy.

Fig. 10. ATVA with variable-curvature piezo-actuated beams (Bonello et al., 2005)

Fig. 11. Servo-actuated beam ATVA: moveable-supports, conventional design (Hong &
Ryu, 1985)


11
In all devices of Figs. 11-13, the actuator is a permanently redundant mass that degrades the
attenuating capacity of the ATVA (eq. (7, 8)). This limitation was overcome by (Carneal et al.,
2004) who improved the moveable-supports concept of (Hong & Ryu, 1985) by
incorporating the actuator into the central mass supported by the beam (Fig. 14). This
necessitated the use of a V-Type undercarriage. The design of this undercarriage (and the
one used in Fig. 11) clearly warrants careful consideration. It should be sufficiently rigid so as
to avoid introducing unwanted dynamics that would interfere with ATVA operation.
Moreover, it needs to be as light as possible to minimise the redundant mass. A far simpler
approach would be to utilise a moveable-masses ATVA with actuators incorporated into the
masses, as illustrated in Fig. 15, where the device simply attaches directly to the host structure
at its centre. Such a novel device was proposed by (Bonello & Groves, 2009). Apart from the
constructional simplicity, this concept was shown to provide superior ATVA performance.
Another important contribution of (Bonello & Groves, 2009) was the derivation of the
effective mass and tuned frequency characteristics of the moveable-supports and moveable-
masses ATVA. This enables the designer to quantify their expected performance for any
given application. The derivation of the effective mass of the beam-like ATVAs in Figs. 11-15
requires the derivation of their equivalent two-degree-of-freedom model (Fig. 1b). Such
analysis is very important when one considers that, for the devices in Fig. 11-15 (with the
possible exception of Fig. 12), the effective mass proportion R will vary as the ATVA is
retuned. Although (Carneal et al., 2004) describe their actuator-incorporated mass (Fig. 14)
as the active mass of the absorber, its degree of activity is actually dependent on the
setting of the ATVA. The same can be said of the moveable-masses concept (Fig. 15). With
the supports or masses fully retracted in Figs. 14 and 15, the attached masses clearly become
entirely redundant and the effective mass proportion R in Fig. 1b is then entirely contributed
by the beam itself. This means that, as the ATVA retunes itself, the attenuation it provides
will vary due to the consequent variation in (eqs. (7, 8)). Hence, the knowledge of an
effective mass characteristic of a moveable-supports or moveable-masses ATVA is
important since it would allow the designer to quantify the expected attenuation provided
by an ATVA over a range of frequencies for any given application.
4. ATVA analysis
The aims of this section are two-fold: (i) to illustrate the derivation of the effective mass and
tuned frequency characteristics of moveable-supports and moveable-masses ATVAs (Figs.
14 and 15); (ii) to illustrate the physical implementation and testing of the beam-like ATVA
with actuator-incorporated moveable masses (Fig. 15). This latter covers the adaptation logic
control. The material in this section is based on the work in (Bonello & Groves, 2009), from
which further details can be obtained.
4.1 Effective mass and tuned frequency characteristics
Fig. 16 shows the two alternative types of ATVA considered. It shall be assumed that the
beam supports of the device in Fig. 16b are simple supports. Let A denote the point/points
of attachment of the beam to the host structure and B denote the point/points of attachment
of the mass/masses. The aim of the following analysis is the determination of the fractional
change in tuned frequency
a
and the variation of the effective mass proportion R (Fig. 1b)
as the setting x x L = of the actual systems in Fig. 16 is varied. Hence, for this purpose,
damping can be omitted from the analysis without loss of accuracy.


12

Fig. 12. Servo-actuated beam-like ATVA: adjustable beam-cross-section ATVA (Brennan,
2000, Kidner et al., 2002)

Fig. 13. Servo-actuated beam-like ATVA: moveable-beam or moveable-masses ATVA
(Brennan, 2000)

Fig. 14. Servo-actuated beam-like ATVA: moveable supports, V-Type design ATVA
(Carneal et al., 2004)


13

Fig. 15. Servo-actuated beam-like ATVA: actuator-incorporated moveable masses ATVA
(Bonello & Groves, 2009)

Fig. 16. Free-body schematics of two alternative designs for an actuator-incorporated mass
ATVA
In either case the system will be regarded as comprising a beam acted upon by: (i) the
reaction forces from the lumped mass attachments at B; (ii) the reaction force from the host.
If the latter force is ( )
{ }
j
Re e
t
f t F

= , then the response at any point Q on the beam is
( )
{ }
j
Re e
t
Q Q
y t Y

= . Following the analysis in (Bonello & Groves, 2009), the expression for
the receptance of the TVA at its point of attachment to the structure is:

( )
( ) ( ) ( ) ( )
( )
{ }
( )
{ }
2
2
1
2 2
2 2
2
2
1 2 2 2
2 2
3,5,....
1
1 1
1
1 1
A
AA
b
AA BB BA AB
k
K
B
B
k k
Y
r
F m

= =

+ + + + +

+

(9)
A
B
B
( ) t f
B
A A
( ) 2 t f
( ) 2 t f
L 2
x 2
L 2
x 2
(a) Moveable masses
(b) Moveable supports (V-type)


14
where:

att
b
m
m
= ,
1
= ,
1
k
k

= , ( )
( ) ( )
2 2
1,3,5,....
k k
K
Q S
QS
k k

=
=

(10a-d)
In the above expressions
att
m is the sum of the attached masses at B (for moveable masses)
or simply the mass at B (for moveable supports) and
b
m is the mass of the beam alone.
( ) k
Q
,
( ) k
S
are the k
th
flexural non-dimensional mode-shape functions evaluated at arbitrary points
Q, S and
k
is the k
th
flexural natural frequency. These modal parameters pertain to the
plain beam (i.e. without lumped mass attachment) under free-free conditions. Notice that
only the symmetric modes are considered due to the symmetry of the systems in Fig. 16. The
expression in eq. (9) is exact for K but, for a given upper limit of the non-dimensional
excitation frequency , will be virtually exact for a sufficiently large finite value of K
(corresponding to a total of ( ) 1 2 K + symmetric modes). For the moveable masses
absorber,
( ) k
A
is fixed and
( ) k
B
is variable; for the moveable supports absorber
( ) k
B
is fixed
and
( ) k
A
is variable. If the Euler-Bernoulli beam model is assumed, then the free-free mode-
shape function values tabulated by (Bishop & Johnson, 1960) can be used. In this case, from
eq. (9), it is clear that the non-dimensional receptance function
( ) ( )
2
1
, ,
AA b AA
r x m r = (11)
is totally independent of the actual material and geometrical properties for either ATVA in
Fig. 16.
For given x and , the lowest non-dimensional anti-resonance
1 a a
= (i.e. the tuned
frequency of the device) is obtained by finding the lowest value of for which the
numerator of eq. (9) is 0:

( ) ( ) ( ) ( )
2 2
2 2
1 1
1 0
AA BB BA AB

+ + + + + =

(12)
Similarly, the lowest non-dimensional resonance
1 m m
= is obtained by finding the
lowest value of for which the denominator of eq. (9) is 0:

( )
{ }
( )
{ }
2
2
1 2 2 2
2 2
3,5,....
1 1 0
k
K
B
B
k k

=

+ =

(13)
For any given K, the frequencies
a
,
m
can be found by an iteration technique using as
initial approximations the corresponding roots
1
a
K
=
,
1
m
K
=
obtained for 1 K = . From eqs.
(12, 13) it can be shown that:

( )
( )
1
2 2
1
1 1 2
a
K
A B A B

=
=
+ + +
,
( )
1 2
1
1 1
m
K
B

=
+
=
+ +
(14a,b)


15
High accuracy is guaranteed here by solving the frequencies
a
,
m
for 39 K = , which
means that 20 symmetric free-free plain beam flexural modes were considered.
Now, for the equivalent two-degree-of-freedom system in Fig. 1b, one can write the
attachment point receptance as (Kidner & Brennan, 1997):
( )
( ) { }
2 2
2-DOF 2 2 2
, , ,
1
a
AA
a red a eff a red a
r
m m m

=
+
(15)
where

, a eff a
m Rm = , ( )
,
1
a red a
m R m = , ( ) 1
a b
m m = + (16a-c)
The non-zero resonance of the function in eq. (15) is given by:

, ,
2-DOF
1
m a a eff a red
m m = + (17)
For equivalence,
2-DOF
m m
= in eq. (17). Hence, by substituting this condition and eqs.
(16a,b) into eq. (17), an expression can be obtained for the proportion R of the total absorber
mass
a
m that is effective in vibration attenuation:
( ) ( )
2
, 1
a m
R x = (18)
where
a
,
m
are the roots of eqs. (12, 13). Also, from eq. (15) and eqs. (16), the non-
dimensional attachment point receptance of the equivalent system can be written as:
( ) ( )
( ) ( )
{ }
2 2
2
1
2 2 2 2-DOF 2-DOF
, ,
1 1
a
AA b AA
a
r x m r
R

= =
+

(19)
The equivalent two-degree-of-freedom model is verified in Fig. 17 against the exact theory
governing the actual (continuous) ATVA structures of Fig. 16 for 5 = and two given
settings 0.25 x = , 0.5 . For each setting of x the corresponding values of
a
and R were
calculated using eqs. (12, 13, 18) and used to plot the function ( )
2-DOF
, ,
AA
r x in eq. (19).
Comparison with the exact receptance ( ) , ,
AA
r x (computed from eqs. (9) and (11)) shows
that the equivalent two-degree-of-freedom system is a satisfactory representation of the
actual systems in Fig. 16 over a frequency range which contains the operational frequency of
the ATVA (
a
= ).
Next, using eqs. (12, 13, 18), the variations of
a
and R with ATVA setting x for various
fixed values of are investigated for both types of ATVA in Fig. 16. The resulting
characteristics are depicted in Fig. 18. With reference to Fig. 18a, it is evident that, as is
increased, the tuning frequency characteristics of both types of ATVA approach each other.
Moreover, for 1 , both types of ATVA give roughly the same overall useable variation in
a
relative to
1
a
x
=
. The moveable-supports ATVA characteristics in Fig. 18a have a peak
(which is more prominent for lower values) that gives the impression of a greater
variation in
a
than the moveable-masses ATVA. However, this is a red herring since


16
these peaks coincide with a stark dip to zero in the effective mass proportion R of the
moveable-supports ATVA, as can be seen in Fig. 18b. These troughs in R are explained by
the fact that, for given , the free body resonance
m
of the moveable-supports ATVA (i.e.
the resonance of the free-free beam with central mass attached) is fixed (i.e. independent of
x ), as can be seen from Figs. 17c,d. Hence, the nodes of the associated mode-shape are fixed
in position so that when the setting x is such that the attachment points A of the moveable-
supports ATVA coincide with these nodes, this ATVA becomes totally useless (i.e.
attenuation 0 D= , eq. (7)).

Fig. 17. Verification of equivalent two-degree-of-freedom model - non-dimensional
attachment point receptance plotted against non-dimensional excitation frequency for two
settings of the ATVAs in Figure 3 with 5 = : exact, through eqs. (9) and (11) ();
equivalent 2-degree-of-freedom model, from eq. (19 ) ()
The moveable-masses ATVA does not suffer from this problem, and consequently has vastly
superior effective mass characteristics, as evident from Fig. 18b. From eqs. (16a, b), one can
rewrite the attenuation D in eq. (8) as:


17

1
1
a a
R
D
R M m

=

+

(20)
It is evident from Fig. 18b and eq. (20) that the degree of attenuation D provided by a given
moveable-supports ATVA in any given application undergoes considerable variability over
its tuning frequency range, dipping to zero at a critical tuned frequency. On the other hand,
the moveable-masses ATVA can be tuned over a comparable tuning frequency range while
providing significantly superior vibration attenuation.

Fig. 18. Tuned frequency and effective mass characteristics for moveable-masses ATVA


18
4.2 Physical implementation and testing
Fig. 19 shows the moveable-masses ATVA with motor-incorporated masses that was built in
(Bonello & Groves, 2009) to lend validation to the theory of the previous section and
demonstrate the ATVA operation. The stepper-motors were operated from the same driver
circuit board through a distribution box that sent identical signals to the motors, ensuring
symmetrically-disposed movement of the masses. Each motor had an internal rotating nut
that moved it along a fixed lead-screw. Each motor was guided by a pair of aluminium
guide-shafts that, along with the lead-screw, made up the beam section.
The aim of Section 4.2.1 is to validate the theory of Section 4.1 whereas the aim of Section
4.2.2 is to demonstrate real-time ATVA operation.
4.2.1 Tuned frequency and effective mass characteristics
In these tests a random signal v was sent to the electrodynamic shaker amplifier and for
each fixed setting x the frequency response function (FRF)
Av
H relating
A
y to v , and the
FRF
BA
T relating
B
y to
A
y (i.e. the transmissibility) were measured. Fig. 20a shows
Av
H
for different settings. The tuned frequency
a
of the ATVA is the anti-resonance, which
coincides with the resonance in
BA
T . Fig. 20b shows that, at the anti-resonance, the cosine of
the phase of
BA
T is approximately zero. This is an indication that the absorber damping
a

(Fig. 1b) is low (Kidner et al., 2002). Hence, just like other types of ATVA e.g. (Rustighi et.
al., 2005, Bonello et al., 2005, Kidner et al., 2002), the cosine of the phase between the
signals
A
y and
B
y can be used as the error signal of a feedback control system for the
ATVA under variable frequency harmonic excitation (Section 4.2.2). It is noted that this
result is in accordance with the two-degree-of-freedom modal reduction of the ATVA and,
additionally, it could be shown theoretically that the cosine of the phase between
A
y and
the signal
Q
y at any other arbitrary point Q on the ATVA would also be zero in the tuned
condition.

Fig. 19. Moveable-masses ATVA demonstrator mounted on electrodynamic shaker (inset
shows motor-incorporated mass and ATVA beam cross-section)


19
Using the FRFs of Fig. 20a and a lumped parameter model of the ATVA/shaker
combination it was possible to estimate the effective mass proportion R of the ATVA for
each setting x , using the analysis described in (Bonello & Groves, 2009). The estimates
varied slightly according to the type of damping assumed for the shaker armature
suspension. However, as can be seen in Fig. 21, regardless of the damping assumption, there
is good correlation with the effective mass characteristic predicted according to the theory of
the previous section. Fig. 22 shows the predicted and measured tuning frequency
characteristic, which gives the ratio of the tuned frequency to the tuned frequency at a
reference setting. The demonstrator did not manage to achieve the predicted 418 % increase
in tuned frequency, although it managed a 255 % increase, which is far higher than other
proposed ATVAs e.g. (Rustighi et. al., 2005, Bonello et al., 2005, Kidner et al., 2002) and
similar to the percentage increase achieved by the V-Type ATVA in (Carneal et al., 2004).
The main reasons for a lower-than-predicted tuned frequency as x was reduced can be
listed as follows: (a) the guide-shafts-pair and lead-screw constituting the beam cross-
section (Fig. 19) would only really vibrate together as one composite fixed-cross-section
beam in bending, as assumed in the theory, if their cross-sections were rigidly secured
relative to each other at regular intervals over the entire beam length this was not the case
in the real system and indeed was not feasible; (b) shear deformation effects induced by the
inertia of the attached masses at B and the reaction force at A became more pronounced as
x was reduced; (c) the slight clearance within the stepper-motors. It is noted that the
limitation in (a) was exacerbated by the offset of the centroidal axis of the lead-screw from
that of the guide-shafts (inset of Fig. 19). Moreover, the limitations described in (a) and (b)
are also encountered when implementing the moveable-supports design (Fig. 16b). It is also
interesting to observe that, at least for the case studied, the divergence in Fig. 22 did not
significantly affect the good correlation in Fig. 21.
4.2.2 Vibration control tests
Fig. 23 shows the experimental set-up for the vibration control tests. The shaker amplifier
was fed with a harmonic excitation signal v of time-varying circular frequency and fixed
amplitude and the ability of the ATVA to attenuate the vibration
A
y by maintaining the
tuned condition
a
= in real time was assessed. The frequency variation occurred over the
interval
i f
t t t < < and was linear:

( ) ( )
( )

i
i
i f i f i i i f
f
f
t t
t t t t t t t
t t

= + < <
(21)
where
i
,
f
are the initial and final frequency values. The swept-sine excitation signal
was hence as given by:
sin v V = ,
d
dt
= (22a,b)


20
where, by substitution of (21) into (22b) and integration:

( ) ( )
( )
( ) ( )
2

0.5
0.5
i
i
f i f i i i i f
f
f f i f i
t
t t
t t t t t t t t
t t
t t t

= + < <

(23)

Fig. 20. Frequency response function measurements for different settings of ATVA of Fig. 19
The inputs to the controller were the signals
A
y ,
B
y from the accelerometers. As discussed
in Section 4.2.1, the instantaneous error signal fed into the controller was ( ) cos e t = and
this was continuously evaluated from
A
y ,
B
y by integrating their normalised product over
a sliding interval of fixed length
c
T , according to the following formula:


21

( )
( )
( ) { } ( ) { }
( )
( ) ( )
( ) ( ) { } ( ) ( ) { }
( )
0.5 0.5
0.5 0.5

cos

AB
c
AA BB
AB AB c
c
AA AA c BB BB c
I t
t T
I t I t
e t
I t I t T
t T
I t I t T I t I t T
= =

>
(24)
where
( ) ( ) ( )
0
t
AA A A
I t y y d =
, ( ) ( ) ( )
0
t
BB B B
I t y y d =
, ( ) ( ) ( )
0
t
AB A B
I t y y d =
(25a-c)
and
c
T was taken to be many times the interval between consecutive sampling times of
the data acquisition ( 100
c
T = typically). Since the difference between the forcing
frequency and the tuned frequency,
a
is non-linearly related to ( ) e t , a non-linear
control algorithm was necessary to minimise ( ) e t . Various such control algorithms for
ATVAs have been proposed. For example, (Bonello et al., 2005) used a nonlinear P-D
controller in which the voltage that controlled the piezo-actuators (Fig. 10) was updated
according to a sum of two polynomial functions, one in e and the other in e , weighted by
suitably chosen constants P and D. (Kidner et al., 2002) formulated a fuzzy logic algorithm
based on e to control the servo-motor of the device in Figure 12b. These algorithms were
not convenient for the present application since they provided an analogue command signal
to the actuator. In the present case, the available motor driver was far more easily operated
through logic signals. Each motor had five possible motion states, respectively activated by
five possible logic-combination inputs to the driver. Hence, the interval-based control
methodology described in Table 1 was implemented, where the error signal computed by
eq. (24) was divided into 5 intervals.

Fig. 21. Effective mass characteristics for prototype moveable masses ATVA: predicted
(); measured, light damping assumption (); measured, proportional damping
assumption ()


22

Fig. 22. Tuned frequency characteristic for prototype moveable masses ATVA: predicted
(); measured ()

Fig. 23. Experimental set-up for vibration control test
The control system for the experimental set-up of Figure 23 was implemented in MATLAB

with SIMULINK
using the Real Time Workshop
and Real Time Windows Target

toolboxes.
Fig. 24 shows the results obtained for the frequency-sweep in Fig. 24a with the control
system turned off and the ATVA tuned to a frequency of 56Hz. It is clear that at the instant
input/output
motor
driver
electrodynamic
shaker
PC running
Simulink
variable frequency
harmonic excitation
signal
B
y
logic output from Simulink
controller
distribution
box
A
y
accelerometers
mass incorporating
stepper motor

B
B
A
amplifier
amplifier


23
where the excitation frequency sweeps through 56 Hz, the amplitude of the acceleration
A
y
is at a minimum value and cos is approximately equal to zero (i.e. the ATVA is
momentarily tuned to the excitation).

Fig. 24. Swept-sine test with controller turned off and ATVA tuned to a fixed frequency of
56 Hz (
peak
A
y
is the amplitude of
A
y
, the tuned acceleration at A at an excitation

frequency of 56 Hz)


24
cos Motion State
1
cos 1 c
Fast CW
2 1
cos c c <
Slow CW
2 2
cos c c < <
Stopped
2 1
cos c c >
Slow CCW
1
cos 1 c
Fast CCW
Table 1. Interval-based control methodology for stepper-motor driver (CW: clockwise;
CCW: counter-clockwise)

Fig. 25. Response to swept sine excitation (Figure 24a) with controller turned on and ATVA
initially tuned to the excitation frequency (controller parameters in Table 1 are
1
0.04 c = , 02 . 0
2
= c ).
Fig. 25 shows the response to the same frequency-sweep of Fig. 24a with the controller turned
on. Prior to the start of the frequency-sweep at 10 t = , the ATVA was allowed to tune itself,
from whatever initial setting it had, to the prevailing excitation frequency of 30 Hz. As the
sweep progressed, the controller retuned the ATVA accordingly to reasonable accuracy, as
illustrated in Fig. 25b. This resulted in minimised vibration over the entire sweep, as evident
by comparing the scales of the vertical axes of Fig. 25a and 24b. However, it is evident from


25
Fig. 25a that the amplitude of the tuned vibration increases steadily over the frequency sweep
between start and finish. Further studies revealed that this observed degradation in
attenuation produced by the ATVA was due to the reduction in its effective mass proportion R
as it retuned itself to a higher frequency, decreasing the effective mass ratio of the ATVA-
shaker combination. This illustrated the importance of knowing the effective mass
characteristic of a moveable-masses or moveable-supports ATVA. It is noted that the tests in
this subsection (4.2.2) were made with an earlier version of the prototype wherein the ATVA
beam came in two halves i.e. one separate lead-screw and a separate guide-shaft-pair for each
symmetric half of the ATVA, each secured into the central block (see Fig. 19). The tests in
section 4.2.1 were made with an improved version wherein the ATVA beam was one
continuous piece, as in the theory (Fig. 16a) i.e. one long lead-screw and guide-shaft pair
running straight through the central block, where they were tightly secured, ensuring a
horizontal slope (see Fig. 19). Based on the validated results of Fig. 21, the observed
degradation in attenuation in Fig. 25a is expected to be much less for the improved version.
5. Conclusion
This chapter started with a quantitative illustration of the basic design principles of both
variants of the TVA: the TMD and the TVN. The importance of adaptive technology,
particularly with regard to the TVN, was justified. The remainder of the chapter then
focussed on adaptive (smart) technology as applied to the TVN. A comprehensive review of
the various design concepts that have been proposed for the ATVA was presented. The
latest ATVA concept introduced by the author, involving a beam-like ATVA with actuator-
incorporated moveable masses, was then studied theoretically and experimentally. The
variation in tuned frequency was shown to be significantly higher than most other proposed
ATVAs and at least as high as that reported in the literature for the alternative moveable-
supports beam ATVA design. Moreover, the analysis revealed that the moveable-masses
beam concept offers significantly superior vibration attenuation relative to the moveable-
supports beam concept, apart from constructional simplicity. Vibration control tests with
logic-based feedback control demonstrated the efficacy of the device under variable
frequency excitation. Current efforts by the author are being directed at introducing smart
technology to TMDs.
6. References
Bishop, R.E.D. & Johnson, D.C. (1960). The Mechanics of Vibration, Cambridge University
Press, Cambridge, UK
Bonello, P. & Brennan, M. J. (2001). Modelling the dynamic behaviour of a supercritical rotor
on a flexible foundation using the mechanical impedance technique. J. Sound and
Vibration, Vol.239, No.3, pp. 445-466
Bonello, P.; Brennan, M. J. & Elliott, S. J. (2005). Vibration control using an adaptive tuned
vibration absorber with a variable curvature stiffness element. Smart Mater. Struct.,
Vol.14, No.5, pp. 1055-1065
Bonello, P. & Groves, K.H. (2009). Vibration control using a beam-like adaptive tuned
vibration absorber with actuator-incorporated mass-element. Proceedings of the
Institution of Mechanical Engineers - Part C: Journal of Mechanical Engineering
Science.,Vol.223.,No.7, pp 1555-1567


26
Brennan, M.J. (1997). Vibration control using a tunable vibration neutraliser. Proc. IMechE
Part C, Journal of Mechanical Engineering Science, Vol.211, pp. 91-108
Brennan, M.J. (2000). Actuators for active control tunable resonant devices. Applied
Mechanics and Engineering, Vol.5, No.1, pp. 63-74
Brennan, M.J.; Bonello, P.; Rustighi, E., Mace, B.R. & Elliott, S.J. (2004a). Designs of a variable
stiffness element for a tunable vibration absorber, Proceedings of ICA2004 (The 18th
International Congress on Acoustics), Vol.IV, pp. 2915-2918, Kyoto, Japan, 4-9
April , 2004
Brennan, M.J.; Bonello, P.; Rustighi, E., Mace, B.R. & Elliott, S.J. (2004b). Designs of a
variable stiffness element for a tunable vibration absorber, Presentation given at
ICA2004 (The 18th International Congress on Acoustics), Vol.IV, pp. 2915-2918, Kyoto,
Japan, 4-9 April , 2004
Carneal, J.P.; Charette, F. & Fuller, C.R. (2004). Minimization of sound radiation from plates
using adaptive tuned vibration absorbers. J. Sound and Vibration, Vol.270, pp.
781-792
Den Hartog, J.P. (1956). Mechanical Vibrations, McGraw Hill (4
th
Edition), New York, USA
Ewins, D.J. (1984). Modal Testing: Theory and Practice, Letchworth: Research Student
Press, UK
Hong, D.P. & Ryu, Y.S. (1985). Automatically controlled vibration absorber. US Patent No.
4935651
Kidner, M.R.F. & Brennan, M.J. (1999). Improving the performance of a vibration neutraliser
by actively removing damping. J. Sound and Vibration, Vo.221, No.4, pp. 587-606
Kidner, M. R. F. & Brennan, M. J. (2002). Variable stiffness of a beam-like neutraliser under
fuzzy logic control. Trans. of the ASME, J. Vibration and Acoustics, Vol.124, pp. 90-99
Ormondroyd, J. & den Hartog, J.P. (1928). Theory of the dynamic absorber. Trans. of the
ASME, Vol. 50, pp. 9-22
Long, T.; Brennan, M.J. & Elliott, S.J. (1998). Design of smart machinery installations to
reduce transmitted vibrations by adaptive modification of internal forces.
Proceedings of the Institution of Mechanical Engineering - Part I: Journal of Systems and
Control Engineering, Vol.212, No.13, pp. 215-228
Longbottom, C.J.; Day M.J. & Rider, E. (1990). A self tuning vibration absorber. UK Patent
No. GB218957B
Park, C.H. (2003). Dynamics modelling of beams with shunted piezoelectric elements. J.
Sound and Vibration, Vol.268, pp. 115-129
Rustighi, E.; Brennan, M.J. & Mace, B.R. (2005). A shape memory alloy adaptive tuned
vibration absorber: design and implementation. Smart Mater. Struct., Vol.14, No.1,
pp. 1928
von Flotow, A.H.; Beard, A.H. & Bailey, D. (1994). Adaptive tuned vibration absorbers:
tuning laws, tracking agility, sizing and physical implementation, Proc. Noise-Con
94, pp. 81-101, Florida, USA, 1994
Francisco Beltran-Carbajal
1
, Gerardo Silva-Navarro
2
,
Benjamin Vazquez-Gonzalez
1
and Esteban Chavez-Conde
3
1
Universidad Autonoma Metropolitana, Plantel Azcapotzalco, Departamento
de Energia, Mexico, D.F.
2
Centro de Investigacion y de Estudios Avanzados del I.P.N., Departamento de Ingenieria
Electrica, Seccion de Mecatronica, Mexico, D.F.
3
Universidad del Papaloapan, Campus Loma Bonita, Departamento
de Mecatronica, Oaxaca
Mexico
1. Introduction
Many engineering systems undergo undesirable vibrations. Vibration control in mechanical
systems is an important problem by means of which vibrations are suppressed or at least
attenuated. In this direction, the dynamic vibration absorbers have been widely applied in
many practical situations because of their low cost/maintenance, efciency, accuracy and
easy installation (Braun et al., 2001; Preumont, 1993). Some of their applications can be found
in buildings, bridges, civil structures, aircrafts, machine tools and many other engineering
systems (Caetano et al., 2010; Korenev & Reznikov, 1993; Sun et al., 1995; Taniguchi et al.,
2008; Weber & Feltrin, 2010; Yang, 2010).
There are three fundamental control design methodologies for vibration absorbers described
as passive, semi-active and active vibration control. Passive vibration control relies on the
addition of stiffness and damping to the primary system in order to reduce its dynamic
response, and serves for specic excitation frequencies and stable operating conditions,
but is not recommended for variable excitation frequencies and/or parametric uncertainty.
Semiactive vibration control deals with adaptive spring or damper characteristics, which are
tuned according to the operating conditions. Active vibration control achieves better dynamic
performance by adding degrees of freedom to the system and/or controlling actuator forces
depending on feedback and feedforward real-time information of the system, obtained from
sensors. For more details about passive, semiactive and active vibration control we refer to
the books (Braun et al., 2001; Den Hartog, 1934; Fuller et al, 1997; Preumont, 1993).
On the other hand, many dynamical systems exhibit a structural property called differential
atness. This property is equivalent to the existence of a set of independent outputs, called
at outputs and equal in number to the control inputs, which completely parameterizes
every state variable and control input (Fliess et al., 1993; Sira-Ramirez & Agrawal, 2004). By
means of differential atness techniques the analysis and design of a controller is greatly

Design of Active Vibration Absorbers Using
On-Line Estimation of Parameters and Signals

2
2 Vibration Control
simplied. In particular, the combination of differential atness with the control approach
called Generalized Proportional Integral (GPI) control, based on output measurements and
integral reconstructions of the state variables (Fliess et al., 2002), qualies as an adequate
control scheme to achieve the robust asymptotic output tracking and, simultaneously, the
cancellation/attenuation of harmonic vibrations. GPI controllers for design of active vibration
absorbers have been previously addressed in (Beltran et al., 2003). Combinations of GPI
control, sliding modes and on-line algebraic identication of harmonic vibrations for design
of adaptive-like active vibration control schemes have been also proposed in (Beltran et al.,
2010). A GPI control strategy implemented as a classical compensation network for robust
perturbation rejection in mechanical systems has been presented in (Sira-Ramirez et al., 2008).
In this chapter a design approach for active vibration absorption schemes in linear
mass-spring-damper mechanical systems subject to exogenous harmonic vibrations is
presented, which are based on differential atness and GPI control, but taking the advantage
of the interesting energy dissipation properties of passive vibration absorbers. Our design
approach considers a mass-spring active vibration absorber as a dynamic controller, which
can simultaneously be used for vibration attenuation and desired reference trajectory tracking
tasks. The proposed approach allows extending the vibrating energy dissipation property of
a passive vibration absorber for harmonic vibrations of any excitation frequency, by applying
suitable control forces to the vibration absorber. Two different active vibration control schemes
are synthesized, one employing only displacement measurements of the primary system and
other using measurements of the displacement of the primary system as well as information
of the excitation frequency. The algebraic parametric identication methodology reported
by (Fliess & Sira-Ramirez, 2003), which employs differential algebra, module theory and
operational calculus, is applied for the on-line estimation of the parameters associated to the
external harmonic vibrations, using only displacement measurements of the primary system.
Some experimental results on the application of on-line algebraic identication of parameters
and excitation forces in vibrating mechanical systems were presented in (Beltran et al., 2004),
which show their success in practical implementations.
The real-time algebraic identication of the excitation frequency is combined with a certainty
equivalence controller to cancel undesirable harmonic vibrations affecting the primary
mechanical systemas well as to track asymptotically and robustly a specied output reference
trajectory. The adaptive-like control scheme results quite fast and robust against parameter
uncertainty and frequency variations.
The main virtue of the proposed identication and adaptive-like control scheme for
vibrating systems is that only measurements of the transient input/output behavior are
used during the identication process, in contrast to the well-known persisting excitation
condition and complex algorithms required by most of the traditional identication methods
(Isermann & Munchhof, 2011; Ljung, 1987; Soderstrom, 1989). It is important to emphasize
that the proposed results are now possible thanks to the existence of high speed DSP boards
with high computational performance operating at high sampling rates.
Finally, some simulation results are provided to show the robust and efcient performance of
the proposed active vibration control schemes as well as of the proposedidentiers for on-line
estimation of the unknown frequency and amplitude of resonant harmonic vibrations.
28 Vibration Analysis and Control New Trends and Developments
Design of Active Vibration Absorbers Using On-line Estimation of Parameters and Signals 3
2. Vibrating mechanical system
2.1 Mathematical model
Consider the vibrating mechanical system shown in Fig. 1, which consists of an active
undamped dynamic vibration absorber (secondary system) coupled to the perturbed
mechanical system (primary system). The generalized coordinates are the displacements of
both masses, x
1
and x
2
, respectively. In addition, u represents the force control input and
f (t) some harmonic perturbation, possibly unknown. Here m
1
, k
1
and c
1
denote mass, linear
stiffness and linear viscous damping on the primary system, respectively. Similarly, m
2
, k
2
and c
2
denote mass, stiffness and viscous damping of the dynamic vibration absorber. Note
also that, when u 0 the active vibration absorber becomes only a passive vibration absorber.
Active Vibration Absorber
Mechanical System
m
1
f(t) = F
0
sin wt
x
1
c
2
= 0 k
2
m
2
c
1
0
u
x
2
k
1
Fig. 1. Schematic diagram of the vibrating mechanical system with active vibration absorber.
The mathematical model of this two degrees-of-freedomsystem is described by the following
two coupled ordinary differential equations
m
1
x
1
+ c
1
x
1
+ k
1
x
1
+ k
2
(x
1
x
2
) = f (t)
m
2
x
2
+ k
2
(x
2
x
1
) = u(t)
(1)
where f (t) = F
0
sint, with F
0
and denoting the amplitude and frequency of the excitation
force, respecively. In order to simplify the analysis we have assumed that c
1
0.
29 Design of Active Vibration Absorbers Using On-Line Estimation of Parameters and Signals
4 Vibration Control
Dening the state variables as z
1
= x
1
, z
2
= x
1
, z
3
= x
2
and z
4
= x
2
, one obtains the following
state-space description
z
1
= z
2
z
2
=
k
1
+k
2
m
1
z
1
c
1
m
1
z
2
+
k
2
m
1
z
3
+
1
m
1
f (t)
z
3
= z
4
z
4
=
k
2
m
2
z
1
k
2
m
2
z
3
+
1
m
2
u(t)
y = z
1
(2)
It is easy to verify that the system (2) is completely controllable and observable as well as
marginally stable in case of c
1
= 0, f 0 and u 0 (asymptotically stable when c
1
> 0). Note
that, an immediate consequence is that, the output y = z
1
has relative degree 4 with respect to
u and relative degree 2 with respect to f and, therefore, the so-called disturbance decoupling
problem of the perturbation f (t) from the output y = z
1
, using state feedback, is not solvable
(Isidori, 1995).
To cancel the exogenous harmonic vibrations on the primary system, the dynamic vibration
absorber should apply an equivalent force to the primary system, with the same amplitude
but in opposite phase (sign). This means that the vibration energy injected to the primary
system, by the exogenous vibration f (t), is transferred to the vibration absorber through
the coupling elements (i.e., spring k
2
). Of course, this vibration control method is possible
under the assumption of perfect knowledge of the exogenous vibrations and stable operating
conditions (Preumont, 1993).
In this work we will apply the algebraic identication method to estimate the parameters
associated to the harmonic force f (t) and then, propose the design of an active vibration
controller based on state feedback and feedforward information of f (t).
2.2 Passive vibration absorber
It is well known that a passive vibration absorber can only cancel the vibration f (t) affecting
the primary system if and only if the excitation frequency coincides with the uncoupled
natural frequency of the absorber (Den Hartog, 1934), that is,
2
=
k
2
m
2
= (3)
See Fig. 2, where X
1
denotes the steady-state maximum amplitude of x
1
(t) and
st
the
static deection of the primary system under the constant force F
0
. Note, however, that
the interconnection of the passive vibration absorber to the primary system slightly changes
the natural frequencies in both uncoupled subsystems and, hence, when =
2
and close
to those resonant frequencies the amplitudes might be large or theoretically innite. This
situation clearly leads to large displacements and could damage of any physical system.
In what follows we shall use an active vibration absorber based on Generalized PI control
(GPI) to provide some robustness with respect to variations on the excitation frequency ,
uncertain system parameters and initial conditions.
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
0
1
2
3
4
5
w/w
2
|
X
1
/
d
s
t
|
Vibration Cancellation at the
Tuning Frequency of the
Absorber w
2
With Passive Vibration Absorber
Without Vibration Absorber
Fig. 2. Frequency response of the vibrating mechanical system with passive vibration
absorber.
2.3 Differential atness
Because the system (2) is completely controllable from u then, it is differentially at, with
at output given by y = z
1
. Then, all the state variables and the control input can be
differentially parameterized in terms of the at output y and a nite number of its time
derivatives (Fliess et al., 1993; Sira-Ramirez & Agrawal, 2004).
In fact, from y and its time derivatives up to fourth order one can obtain that
y = z
1
y = z
2
y =
k
1
+k
2
m
1
z
1
+
k
2
m
1
z
3
y
(3)
=
k
1
+k
2
m
1
z
2
+
k
2
m
1
z
4
y
(4)
=
_
(k
1
+k
2
)
2
m
2
1
+
k
2
2
m
1
m
2
_
z
1
_
k
2
(k
2
+k
1
)
m
2
1
+
k
2
2
m
1
m
2
_
z
3
+
k
2
m
1
m
2
u
(4)
where c
1
= 0 and f 0. Therefore, the differential parameterization results as follows
z
1
= y
z
2
= y
z
3
=
k
1
+k
2
k
2
y +
m
1
k
2
y
z
4
=
k
1
+k
2
k
2
y +
m
1
k
2
y
(3)
u = k
1
y +
_
m
1
+ m
2
+
k
1
k
2
m
2
_
y +
m
1
m
2
k
2
y
(4)
(5)
Then, the at output y satises the following input-output differential equation
y
(4)
= a
0
y + a
2
y + bu (6)
6 Vibration Control
where
a
0
=
k
1
k
2
m
1
m
2
a
2
=
_
k
1
+ k
2
m
1
+
k
2
m
2
_
b =
k
2
m
1
m
2
From (6) one obtains the following differential atness-based controller to asymptotically
track some desired reference trajectory y
(t):
u = b
1
(v a
0
y a
2
y) (7)
with
v = (y
)
(4)
(t)
6
_
y
(3)
(y
)
(3)
(t)
_

5
[ y y
(t)]
4
[ y y
(t)]
3
[y y
(t)]
The use of this controller yields the following closed-loop dynamics for the trajectory tracking
error e = y y
(t):
e
(4)
+
6
e
(3)
+
5
e +
4
e +
3
e = 0 (8)
Therefore, selecting the designparameters
i
, i = 3, ..., 6, such that the associated characteristic
polynomial for (8) be Hurwitz, i.e., all its roots lying in the open left half complex plane, one
can guarantee that the error dynamics be globally asymptotically stable.
Nevertheless, this controller is not robust with respect to exogenous signals or parameter
uncertainties in the model. In case of f (t) = 0, the parameterization should explicitly include
the effect of f and its time derivatives up to second order. In addition, the implementation
of this controller requires measurements of the time derivatives of the at output up to third
order and vibration signal and its time derivatives up to second order.
Remark. In spite of the linear models under study, it results important to emphasize the great
potential of the differential atness approach for nonlinear at systems, which can be analyzed
using similar arguments (Fliess & Sira-Ramirez, 2003). In fact, the proposed results can be
generalized to some classes of nonlinear mechanical systems.
Next, we will synthesize two controllers based on the Generalized PI (GPI) control approach
combined with differential atness and passive absorption, in order to get robust controllers
against external vibrations.
3. Generalized PI control
3.1 Control scheme using displacement measurement on the primary system
Since the system (2) is observable for the at output y then, all the time derivatives of the at
output can be reconstructed by means of integrators, that is, they can be expressed in terms
of the at output y, the input u and iterated integrals of the input and the output variables
(Fliess et al., 2002).
For simplicity, we will denote the integral
_
t
0
() d by
_
and
_
t
0
_
1
0

_
n1
0
(
n
) d
n
d
1
by
_
(n)
with n a positive integer. The integral input-output
parameterization of the time derivatives of the at output is given, modulo initial conditions,
by
y = a
0
_
(3)
y + a
2
_
y + b
_
(3)
u
y = a
0
_
(2)
y + a
2
y + b
_
(2)
u
y
(3)
= a
0
_
y + a
2
y + b
_
u
(9)
These expressions were obtained by successive integrations of the last equation in (6). For
non-zero initial conditions, the relations linking the actual values of the time derivatives of
the at output to the structural estimates in (9) are given as follows
y =

y + e
12
t
2
+ e
11
t + e
11
y =

y + g
11
t + g
10
y
(3)
=

y
(3)
+ h
12
t
2
+ h
11
t + h
10
(10)
where e
1i
, g
j
, h
i
, i = 0, . . . , 2, j = 0, . . . , 1, are real constants depending on the unknown initial
conditions.
For the design of the GPI controller, the time derivatives of the at output are replacedfor their
structural estimates (9) into (7). This, however, implies that the closed-loop system would
be actually excited by constant values, ramps and quadratic functions. To eliminate these
destabilizing effects of such structural estimation errors, one can use the following controller
with iterated integral error compensation:
u = b
1
_
v a
0
y a
2
y
_
v = (y
)
(4)
(t)
6
_
y
(3)
(y
)
(3)
(t)
_

5
_
y y
(t)

4
_
y y
(t)
3
[y y
(t)]
2
1

1
2

0
1
= y y
(t) ,
1
(0) = 0
2
=
1
,
2
(0) = 0
3
=
2
,
3
(0) = 0
(11)
The use of this controller yields the following closed-loop system dynamics for the tracking
error, e = y y
(t), described by
e
(7)
+
6
e
(6)
+
5
e
(5)
+
4
e
(4)
+
3
e
(3)
+
2
e +
1
e +
0
e = 0 (12)
The coefcients
i
, i = 0, ..., 6, have to be selected in such way that the characteristic
polynomial of (12) be Hurwitz. Thus, one can conclude that lim
t
e (t) = 0, i.e., the asymptotic
output tracking of the reference trajectory lim
t
y (t) = y
(t).
3.1.1 Robustness analysis with respect to external vibrations
Now, consider that the passive vibration absorber is tuned at the uncoupled natural frequency
of the primary system, that is,
2
=
1
. The transfer function of the closed-loop system from
8 Vibration Control
the perturbation f (t) to the output y = z
1
is then given by
G (s) =
x
1
(s)
f (s)
=

_
m
2
s
2
+ k
2
_ _
m
2
s
3
+
6
m
2
s
2
+
5
m
2
s
6
k
2
2
6
k
2
+
4
m
2
2k
2
s k
2
s
_
m
3
2
_
s
7
+
6
s
6
+
5
s
5
+
4
s
4
+
3
s
3
+
2
s
2
+
1
s +
0
_
(13)
where = m
2
/m
1
is the mass ratio.
Then, for the harmonic perturbation f (t) = F
0
sint, the steady-state magnitude of the
primary system is computed as
|X
1
| =

m
3
2
F
0
A()
B()
(14)
where
A() =
_
k
2
m
2
2
_
2
_
_
6
m
2
2

6
k
2
2
6
k
2
+
4
m
2
_
2
+
_
m
2
3
+
5
m
2
2k
2
k
2
_
2
_
B() =
_
6
+
4
4

2
2
+
0
_
2
+
_
7
+
5
5

3
3
+
1
_
2
Note that X
1
0 exactly when =
2
=
_
k
2
m
2
, independently of the selected gains of the
control lawin (11), corresponding to the dynamic performance of the passive vibration control
scheme. This clearly corresponds to a nite zero in the above transfer function G(s), situation
where the passive vibration absorber is well tuned.
Thus, the control objective for (11) is to add some robustness when =
2
and improve the
performance of the closed-loop systemusing small control efforts and taking advantage of the
passive vibration absorber (when =
2
the system can operate with u 0).
In Fig. 3 we can observe that, the active vibration absorber can attenuate vibrations for any
excitation frequency, including vibrations with multiple harmonic signals. In fact, it is still
possible to minimize the attenuation level by adding a proper viscous damping to the absorber
(Korenev & Reznikov, 1993; Rao, 1995).
3.2 Control scheme using displacement measurement on the primary system and
excitation frequency
Consider the perturbed system (2). The state variables and the control input u can be
expressed in terms of the at output y, the perturbation f and their time derivatives:
z
1
= y
z
2
= y
z
3
=
k
1
+k
2
k
2
y +
m
1
k
2
y
1
k
2
f (t)
z
4
=
k
1
+k
2
k
2
y +
m
1
k
2
y
(3)
1
k
2
f (t)
u =
m
1
m
2
k
2
y
(4)
+ k
1
y +
_
m
1
+ m
2
+
k
1
k
2
m
2
_
y f (t)
m
2
k
2
f (t)
(15)
0 0.5 1 1.5 2 2.5
0
1
2
3
4
5
6
|
X
1
/
d
s
t
|
w/w
2
Absorber w
2
Fig. 3. Frequency response of the vibrating mechanical system using an active vibration
absorber with controller (11).
Furthermore, when f (t) = F
0
sint the at output y satises the following input-output
differential equation:
y
(4)
=
k
1
k
2
m
1
m
2
y
_
k
1
+ k
2
m
1
+
k
2
m
2
_
y +
_
k
2
m
1
m
2

2
m
1
_
F
0
sint +
k
2
m
1
m
2
u (16)
Taking two additional time derivatives of (16) results in
y
(6)
=
k
1
k
2
m
1
m
2
y
_
k
1
+ k
2
m
1
+
k
2
m
2
_
y
(4)
+
k
2
m
1
m
2
u
_
k
2
m
1
m
2

2
m
1
_
2
F
0
sin t (17)
Multiplication of (16) by
2
and adding it to (17) leads to
y
(6)
+ d
1
y
(4)
+ d
2
y + d
3
y = d
4
_
u +
2
u
_
(18)
where
d
1
=
k
1
+k
2
m
1
+
k
2
m
2
+
2
d
2
=
_
k
1
+k
2
m
1
+
k
2
m
2
_
2
+
k
1
k
2
m
1
m
2
d
3
=
k
1
k
2
m
1
m
2
2
d
4
=
k
2
m
1
m
2
A differential atness-based dynamic controller, using feedback measurements of the at
output y and its time derivatives up to fth order as well as feedforward measurements of
the excitation frequency , is proposed by the following dynamic compensator:
u +
2
u = d
1
4
v + d
1
4
_
d
1
y
(4)
+ d
2
y +
k
1
k
2
m
1
m
2
2
y
_
v = y
(6)
10
_
y
(5)
y
(5)
_
9
_
y
(4)
y
(4)
_
8
_
y
(3)
y
(3)
_
7
[ y y
]
6
[ y y
]
5
[y y
]
(19)
10 Vibration Control
with zero initial conditions (i.e., u(0) = u(0) = 0). It is important to remark that, the above
differential equation resembles an exosystem (linear oscillator) tuned at the known excitation
frequency (feedforward action) and injected by feedback terms involving the at output y
and its desired reference trajectory y
.
On the other hand, one can note that the time derivatives of the at output admit an integral
input-output parameterization, obtained after some algebraic manipulations, given by
y = d
1
_
y d
2
_
(3)
y d
3
_
(5)
y + d
4
_
(3)
u
y = d
1
y d
2
_
(2)
y d
3
_
(4)
y + d
4
_
(2)
u + d
4
2
_
(4)
u
y
(3)
= d
1
y d
2
_
y d
3
_
(3)
y + d
4
_
u + d
4
2
_
(3)
u
y
(4)
= d
1
y d
2
y d
3
_
(2)
y + d
4
u + d
4
2
_
(2)
u
y
(5)
= d
1
y
(3)
d
2
y d
3
_
y + d
4
u + d
4
2
_
u
(20)
The differences in the structural estimates of the time derivatives of the at output with respect
to the actual time derivatives are given by
y =

y + p
4
t
4
+ p
3
t
3
+ p
2
t
2
+ p
1
t + p
0
y =

y + p
4
t
3
+ p
3
t
2
+ p
2
t + p
1
y
(3)
= y
(3)
+ q
4
t
4
+ q
3
t
3
+ q
2
t
2
+ q
1
t + q
0
y
(4)
= y
(4)
+ r
3
t
3
+ r
2
t
2
+ r
1
t + r
0
y
(5)
= y
(5)
+ s
4
t
4
+ s
3
t
3
+ s
2
t
2
+ s
1
t + s
0
where p
i
, q
i
, r
j
, s
i
, i = 0,...,4, j = 0, ..., 3, are real constants depending on the unknown initial
conditions.
Finally, the differential atness based GPI controller is obtained by replacing the actual time
derivatives of the at output in (19) by their structural estimates in (20) but using additional
iterated integral error compensations as follows
u +
2
u = d
1
4
v + d
1
4
_
d
1
y
(4)
+ d
2
y + d
3
y
_
v = y
(6)
10
_
y
(5)
y
(5)
_
9
_
y
(4)
y
(4)
_
8
_
y
(3)
y
(3)
_
7
_
y y
6
_
y y
5
[y y
1
= y y
,
1
(0) = 0
2
=
1
,
2
(0) = 0
3
=
2
,
3
(0) = 0
4
=
3
,
4
(0) = 0
5
=
4
,
5
(0) = 0
(21)
This feedback and feedforward active vibration controller depends on the measurements
of the at output y and the excitation frequency , therefore, this dynamic controller can
compensate simultaneously two harmonic components, corresponding to the tuned (passive)
vibration absorber (
2
) and the actual excitation frequency ().
The closed-loop system dynamics, expressed in terms of the tracking error e = y y
(t), is
described by
e
(11)
+
10
e
(10)
+
9
e
(9)
+
8
e
(8)
+
7
e
(7)
+
6
e
(6)
+
5
e
(5)
+
4
e
(4)
+
3
e
(3)
+
2
e +
1
e +
0
e = 0
(22)
Therefore, the design parameters
i
, i = 0, ..., 10, have to be selected such that the associated
characteristic polynomial for (22) be Hurwitz, thus guaranteeing the desiredasymptotic output
tracking when one can measure the excitation frequency .
3.2.1 Robustness with respect to external vibrations
Fig. 4 shows the frequency response of the closed-loop system, using an active vibration
absorber based on differential atness and measurements of y and . Note that this active
0 0.5 1 1.5 2 2.5
0
1
2
3
4
5
6
7
w/w
2
|
X
1
/
d
s
t
|
Vibration Cancellation at
w
s
/w
1
= 0.8
k
1
= 1000 [N/m]
m
1
= 10 [Kg]
k
2
= 200 [N/m]
m
2
= 2 [Kg]
Absorber w
2
Fig. 4. Frequency response of the vibrating mechanical system using the active vibration
absorber with controller (21).
vibration absorber employs the measurement of the excitation frequency and, therefore,
such harmonic vibrations can always be cancelled (i.e., X
1
= 0). Moreover, this absorber is
also useful to eliminate vibrations of the form f (t) = F
0
[sin (
s
t) +sin (
2
t)], where
s
is the
measured frequency (affecting the feedforward control action) and
2
is the design frequency
of the passive absorber.
3.3 Simulation results
Some numerical simulations were performed on a vibrating mechanical platform from
Educational Control Products (ECP), model 210/210a Rectilinear Control System, characterized by
the set of system parameters given in Table 1.
The controllers (11) and (21) were specied in such a way that one could prove how the active
vibration absorber cancels the two harmonic vibrations affecting the primary system and the
asymptotic output tracking of the desired reference trajectory.
m
1
= 10kg m
2
= 2kg
k
1
= 1000
N
m
k
2
= 200
N
m
c
1
0
N
m/s
c
2
0
N
m/s
Table 1. System parameters for the primary and secondary systems.
The planned trajectory for the at output y = z
1
is given by
y
(t) =
0 for 0 t < T
1
(t, T
1
, T
2
) y for T
1
t T
2
y for t > T
2
where y = 0.01m, T
1
= 5s, T
2
= 10s and (t, T
1
, T
2
) is a Bzier polynomial, with
(T
1
, T
1
, T
2
) = 0 and (T
2
, T
1
, T
2
) = 1, described by
(t) =
_
t T
1
T
2
T
1
_
5
_
r
1
r
2
_
t T
1
T
2
T
1
_
+ r
3
_
t T
1
T
2
T
1
_
2
... r
6
_
t T
1
T
2
T
1
_
5
_
with r
1
= 252, r
2
= 1050, r
3
= 1800, r
4
= 1575, r
5
= 700, r
6
= 126.
Fig. 5 shows the dynamic behavior of the closed-loop system with the controller (11). One
can observe the vibration cancellation on the primary system and the output tracking of the
pre-specied reference trajectory. The controller gains were chosen so that the characteristic
polynomial of the closed-loop tracking error dynamics (12) is a Hurwitz polynomial given by
p
d1
(s) = (s + p
1
)
_
s
2
+2
1
n1
s +
2
n1
_
3
with
1
= 0.5,
n1
= 12rad/s, p
1
= 6.
0 2 4 6 8 10 12
-0.01
-0.005
0
0.005
0.01
0.015
time [s]
z
1
[
m
]
0 2 4 6 8 10 12
-0.04
-0.02
0
0.02
0.04
0.06
0.08
time [s]
z
3
[
m
]
0 2 4 6 8 10 12
-5
0
5
10
15
time [s]
u
[
N
]
Fig. 5. Active vibration absorber using measurements of y and for harmonic vibration
f = 2 sin(10t) N.
Fig. 6 shows the robust performance of closed-loop system employing the controller
(21). One can see that the active vibration absorber eliminates vibrations containing two
different harmonics. The design parameters were selected to have a sixth order closed-loop
characteristic polynomial
p
d2
(s) = (s + p
2
)
_
s
2
+2
2
n2
s +
2
n2
_
5
with
2
= 0.5,
n2
= 10rad/s, p
2
= 5.
0 2 4 6 8 10 12
-0.015
-0.01
-0.005
0
0.005
0.01
0.015
time [s]
z
1
[
m
]
0 2 4 6 8 10 12
-0.1
-0.05
0
0.05
0.1
time [s]
z
3
[
m
]
0 2 4 6 8 10 12
-10
-5
0
5
10
15
time [s]
u
[
N
]
Fig. 6. Active vibration absorber using measurements of y and , for external vibration
f = 2 [sin(8.0109t) +sin (10t)] N.
4. Algebraic identication of harmonic vibrations
The algebraic identication methodology (Fliess & Sira-Ramirez, 2003) can be applied to
estimate the parameters associated to exogenous harmonic vibrations affecting a mechanical
vibrating system (Beltran et al., 2004).
To do that, consider the input-output differential equation (16), where only displacement
measurements of the primary system y = z
1
and the control input u are available for the
identication process of the parameters associated with the harmonic signal f (t) = F
0
sint,
that is,
y
(4)
=
k
1
k
2
m
1
m
2
y
_
k
1
+ k
2
m
1
+
k
2
m
2
_
y +
_
k
2
m
1
m
2

2
m
1
_
F
0
sint +
k
2
m
1
m
2
u (23)
Next we will proceed to synthesize two algebraic identiers for the excitation frequency
and amplitude F
0
.
4.1 Identication of the excitation frequency
The differential equation (23) is expressed in notation of operational calculus as
m
1
s
4
Y (s) +
_
k
1
+ k
2
+
m
1
k
2
m
2
_
s
2
Y (s) +
k
1
k
2
m
2
Y (s)
=
k
2
m
2
U (s) +
_
k
2
m
2
2
_
F
0
s
2
+
2
+ a
3
s
3
+ a
2
s
2
+ a
1
s + a
0
(24)
where a
i
, i = 0,...,3, denote unknown real constants depending on the system initial
conditions. Now, equation (24) is multiplied by
_
s
2
+
2
_
, leading to
_
s
2
+
2
_
__
s
4
Y +
k
2
m
2
s
2
Y
_
m
1
+
_
s
2
Y +
k
2
m
2
Y
_
k
1
+ k
2
s
2
Y
_
=
k
2
m
2
_
s
2
+
2
_
u +
_
k
2
m
2
2
_
F
0
+
_
s
2
+
2
_ _
a
3
s
3
+ a
2
s
2
+ a
1
s + a
0
_
(25)
This equation is then differentiated six times with respect to s, in order to eliminate the
constants a
i
and the unknown amplitude F
0
. The resulting equation is then multiplied by
s
6
to avoid differentiations with respect to time in time domain, and next transformed into
the time domain, to get
_
a
11
(t) +
2
a
12
(t)
_
m
1
+
_
a
12
(t) +
2
b
12
(t)
_
k
1
= c
1
(t) +
2
d
1
(t) (26)
where t = t t
0
and
a
11
(t) = m
2
g
11
(t) + k
2
g
12
(t)
a
12
(t) = m
2
g
12
(t) + k
2
g
13
(t)
b
12
(t) = m
2
g
13
(t) + k
2
_
(6)
t
0
(t)
6
z
1
c
1
(t) = k
2
g
14
(t) k
2
m
2
g
12
(t)
d
1
(t) = k
2
_
(6)
t
0
(t)
6
u k
2
m
2
g
13
(t)
with
g
11
(t) = 720
_
(6)
t
0
y4320
_
(5)
t
0
(t) y+5400
_
(4)
t
0
(t)
2
y2400
_
(3)
t
0
(t)
3
y
+450
_
(2)
t
0
(t)
4
y36
_
t
0
(t)
5
y+ (t)
6
y
g
12
(t) = 360
_
(6)
t
0
(t)
2
y480
_
(5)
t
0
(t)
3
y+180
_
(4)
t
0
(t)
4
y
24
_
(3)
t
0
(t)
5
y+
_
(2)
t
0
(t)
6
y
g
13
(t) = 30
_
(6)
t
0
(t)
4
y12
_
(5)
t
0
(t)
5
y+
_
(4)
t
0
(t)
6
y
g
14
(t) = 30
_
(6)
t
0
(t)
4
u12
_
(5)
t
0
(t)
5
u+
_
(4)
t
0
(t)
6
u
Finally, solving for the excitation frequency in (26) leads to the following on-line algebraic
identier:
2
e
=
N
1
(t)
D
1
(t)
=
c
1
(t) a
11
(t) m
1
a
12
(t) k
1
a
12
(t) m
1
+ b
12
(t) k
1
d
1
(t)
(27)
This estimation is valid if and only if the condition D
1
(t) = 0 holds in a sufciently small time
interval (t
0
, t
0
+
0
] with
0
> 0. This nonsingularity condition is somewhat similar to the
well-known persistent excitation property needed by most of the asymptotic identication
methods (Isermann & Munchhof, 2011; Ljung, 1987; Soderstrom, 1989). In particular, this
obstacle can be overcome by using numerical resetting algorithms or further integrations on
N
1
(t) and D
1
(t) (Sira-Ramirez et al., 2008).
4.2 Identication of the amplitude F
0
To synthesize an algebraic identier for the amplitude F
0
of the harmonic vibrations acting on
the mechanical system, the input-output differential equation (23) is expressed in notation of
operational calculus as follows
m
1
s
4
Y (s) +
_
k
1
+ k
2
+
m
1
k
2
m
2
_
s
2
Y (s) +
k
1
k
2
m
2
Y (s)
=
k
2
m
2
U (s) +
_
k
2
m
2
2
_
F (s) + a
3
s
3
+ a
2
s
2
+ a
1
s + a
0
(28)
Taking derivatives, four times, with respect to s makes possible to remove the dependence
on the unknown constants a
i
. The resulting equation is then multiplied by s
4
, and next
transformed into the time domain, to get
m
1
P
1
(t) +
_
k
1
+ k
2
+
m
1
k
2
m
2
_
P
2
(t) +
k
1
k
2
m
2
_
(4)
t
0
(t)
4
z
1
=
k
2
m
2
_
(4)
t
0
(t)
4
u + F
0
_
k
2
m
2
2
_
_
(4)
t
0
(t)
4
sin t (29)
where
P
1
(t) = 24
_
(4)
t
0
z
1
96
_
(3)
t
0
(t) z
1
+72
_
(2)
t
0
(t)
2
z
1
16
_
t
0
(t)
3
z
1
+ (t)
4
z
1
P
2
(t) = 12
_
(4)
t
0
(t)
2
z
1
8
_
(3)
t
0
(t)
3
z
1
+
_
(2)
t
0
(t)
4
z
1
It is important to note that equation (29) still depends on the excitation frequency , which can
be estimated from (27). Therefore, it is required to synchronize both algebraic identiers for
and F
0
. This procedure is sequentially executed, rst by running the identier for and, after
some small time interval with the estimation
e
(t
0
+
0
) is then started the algebraic identier
for F
0
, which is obtained by solving
N
2
(t) D
2
(t)F
0
= 0 (30)
where
N
2
(t) = m
1
P
1
(t) +
_
k
1
+ k
2
+
m
1
k
2
m
2
_
P
2
(t) +
k
1
k
2
m
2
_
(4)
t
0
+
0
(t)
4
z
1
k
2
m
2
_
(4)
t
0
+
0
(t)
4
u
D
2
(t) =
_
k
2
m
2
2
e
_
_
(4)
t
0
+
0
(t)
4
sin [
e
(t
0
+
0
)t]
In this case if the condition D
2
(t) = 0 is satised for all t (t
0
+
0
, t
0
+
1
] with
1
>
0
> 0,
then the solution of (30) yields an algebraic identier for the excitation amplitude
F
0e
=
N
2
(t)
D
2
(t)
, t (t
0
+
0
, t
0
+
1
] (31)
4.3 Adaptive-like active vibration absorber for unknown harmonic forces
The active vibration control scheme (21), based on the differential atness property and
the GPI controller, can be combined with the on-line algebraic identication of harmonic
vibrations (27) and (31), where the estimated harmonic force is computed as
f
e
(t) = F
0e
sin(
e
t) (32)
resulting some certainty equivalence feedback/feedforward control law. Note that, according
to the algebraic identication approach, providing fast identication for the parameters
associated to the harmonic vibration (F
0
, ) and, as a consequence, a fast estimation of this
perturbation signal, the proposed controller is similar to an adaptive control scheme. From a
theoretical point of view, the algebraic identication is instantaneous (Fliess & Sira-Ramirez,
2003). In practice, however, there are modelling errors and many other factors that
complicate the real-time algebraic computation. Fortunately, the identication algorithms and
closed-loop system are robust against such difculties.
4.4 Simulation results
Fig. 7 shows the identication process of the excitation frequency of the resonant harmonic
perturbation f (t) = 2 sin(8.0109t) N and the robust performance of the adaptive-like control
scheme (21) for reference trajectory tracking tasks, which starts using the nominal frequency
value = 10rad/s, which corresponds to the known design frequency for the passive
vibration absorber, and at t > 0.1s this controller uses the estimated value of the resonant
excitation frequency. Here it is clear how the frequency identication is quickly performed
(before t = 0.1s and it is almost exact with respect to the actual value.
One can also observe that, the resonant vibrations affecting the primary mechanical systemare
asymptotically cancelled from the primary system response in a short time interval. It is also
evident the presence of some singularities in the algebraic identier, i.e., when its denominator
D
1
(t) is zero. The rst singularity, however, occurs about t = 0.727s, which is too much time
(more than 7 times) after the identication has been nished.
Fig. 8 illustrates the fast and effective performance of the on-line algebraic identier for the
amplitude of the harmonic force f (t) = 2 sin (8.0109t) N. First of all, it is started the identier
for , which takes about t < 0.1s to get a good estimation. After the time interval (0, 0.1]s,
where t
0
= 0s and
0
= 0.1s with an estimated value
e
(t
0
+
0
) = 8.0108rad/s, it is activated
the identier for the amplitude F
0
.
0 0.05 0.1 0.15 0.2
0
2
4
6
8
10
time [s]
w
e
[
r
a
d
/
s
]
0 0.25 0.5 0.75
-3
-2
-1
0
1
2
3
x 10
-5
time [s]
0 0.25 0.5 0.75
-6
-4
-2
0
2
4
6
x 10
-7
time [s]
0 5 10 15
-0.02
-0.01
0
0.01
0.02
time [s]
z
1
[
m
]
0 5 10 15
-0.05
0
0.05
0.1
0.15
time [s]
z
3
[
m
]
0 5 10 15
-5
0
5
10
15
time [s]
u
[
N
]
N
1
D
1
Fig. 7. Controlled system responses and identication of frequency for f (t) = 2 sin(8.0109t)
[N].
0 0.05 0.1 0.15 0.2
0
0.5
1
1.5
2
2.5
t [s]
F
0
e
0 0.25 0.5 0.75
-0.5
0
0.5
1
1.5
2
2.5
x 10
-5
t [s]
0 0.25 0.5 0.75
-4
0
4
8
12
x 10
-6
t [s]
N
2
D
2
Fig. 8. Identication of amplitude for f (t) = 2 sin (8.0109t) [N].
One can also observe that the rst singularity occurs when the numerator N
2
(t) and
denominator D
2
(t) are zero. However the rst singularity is presented about t = 0.702s,
and therefore the identication process is not affected.
Now, Figs. 9 and 10 present the robust performance of the on-line algebraic identiers for
the excitation frequency and amplitude F
0
. In this case, the primary system was forced
by external vibrations containing two harmonics, f (t) = 2 [sin (8.0109t) +10 sin(10t)] N.
Here, the frequency
2
= 10rad/s corresponds to the known tuning frequency of the
passive vibration absorber, which does not need to be identied. Once again, one can see
the fast and effective estimation of the resonant excitation frequency = 8.0109rad/s and
amplitude F
0
= 2N as well as the robust performance of the proposed active vibration control
scheme (21) based on differential atness and GPI control, which only requires displacement
measurements of the primary system and information of the estimated excitation frequency.
0 0.05 0.1 0.15 0.2
0
2
4
6
8
10
t [s]
w
e
[
r
a
d
/
s
]
0 0.25 0.5 0.75
-0.5
0
0.5
1
1.5
2
2.5
x 10
-5
t [s]
0 0.25 0.5 0.75
-0.5
0
0.5
1
1.5
2
2.5
3
3.5
x 10
-7
t [s]
0 5 10 15
-0.02
-0.01
0
0.01
0.02
t [s]
z
1
[
m
]
0 5 10 15
-0.1
-0.05
0
0.05
0.1
t [s]
z
3
[
m
]
0 5 10 15
-10
-5
0
5
10
15
t [s]
u
[
N
]
N
1
D
1
Fig. 9. Controlled system responses and identication of the unknown resonant frequency
for . f (t) = 2 [sin (8.0109t) +10 sin(10t)] [N].
0 0.05 0.1 0.15 0.2
0
0.5
1
1.5
2
2.5
t [s]
F
0
e
0 0.25 0.5 0.75
-0.5
0
0.5
1
1.5
2
2.5
x 10
-5
t [s]
0 0.25 0.5 0.75
-2
0
2
4
6
8
10
12
x 10
-6
t [s]
N
2
D
2
Fig. 10. Identication of amplitude for f (t) = 2 [sin (8.0109t) +10 sin(10t)] [N].
5. Conclusions
In this chapter we have described the design approach of a robust active vibration absorption
scheme for vibrating mechanical systems based on passive vibration absorbers, differential
atness, GPI control and on-line algebraic identication of harmonic forces.
The proposed adaptive-like active controller is useful to completely cancel any harmonic
force, with unknown amplitude and excitation frequency, and to improve the robustness
of passive/active vibrations absorbers employing only displacement measurements of the
primary system and small control efforts. In addition, the controller is also able to
asymptotically track some desired reference trajectory for the primary system.
In general, one can conclude that the adaptive-like vibration control scheme results quite
fast and robust in presence of parameter uncertainty and variations on the amplitude and
excitation frequency of harmonic perturbations.
The methodology can be applied to rotor-bearing systems and some classes of nonlinear
mechanical systems.
6. References
Beltran-Carbajal, F., Silva-Navarro, G. & Sira-Ramirez, H. (2003). Active Vibration Absorbers
Using Generalized PI and Sliding-Mode Control Techniques, Proceedings of the
American Control Conference 2003, pp. 791-796, Denver, CO, USA.
Beltran-Carbajal, F., Silva-Navarro, G. & Sira-Ramirez, H. (2004). Application of On-line
Algebraic Identication in Active Vibration Control, Proceedings of the International
Conference on Noise & Vibration Engineering 2004, pp. 157-172, Leuven, Belgium, 2004.
Beltran-Carbajal, F., Silva-Navarro, G., Sira-Ramirez, H., Blanco-Ortega, A. (2010). Active
Vibration Control Using On-line Algebraic Identication and Sliding Modes,
Computacin y Sistemas, Vol. 13, No. 3, pp. 313-330.
Braun, S.G., Ewins, D.J. & Rao, S.S. (2001). Encyclopedia of Vibration, Vols. 1-3, Academic Press,
San Diego, CA.
Caetano, E., Cunha, A., Moutinho, C. & Magalhes, F. (2010). Studies for controlling
human-induced vibration of the Pedro e Ins footbridge, Portugal. Part 2:
Implementation of tuned mass dampers, Engineering Structures, Vol. 32, pp.
1082-1091.
Den Hartog, J.P. (1934). Mechanical Vibrations, McGraw-Hill, NY.
Fliess, M., Lvine, J., Martin, P. & Rouchon, P. (1993). Flatness and defect of nonlinear systems:
Introductory theory and examples, International Journal of Control, Vol. 61(6), pp.
1327-1361.
Fliess, M., Marquez, R., Delaleau, E. & Sira-Ramirez, H. (2002). Correcteurs
Proportionnels-Integraux Gnraliss, ESAIM Control, Optimisation and Calculus
of Variations, Vol. 7, No. 2, pp. 23-41.
Fliess, M. & Sira-Ramirez, H. (2003). An algebraic framework for linear identication, ESAIM:
Control, Optimization and Calculus of Variations, Vol. 9, pp. 151-168.
Fuller, C.R., Elliot, S.J. & Nelson, P.A. (1997). Active Control of Vibration, Academic Press, San
Diego, CA.
Isermann, R. & Munchhpf, M. (2011). Identication of Dynamic Systems, Springer-Verlag, Berlin.
Isidori, A. (1995). Nonlinear Control Systems, Springer-Verlag, NY.
Ljung, L. (1987). Systems Identication: Theory for the User, Prentice-Hall, Englewood Cliffs, NJ.
Korenev, B.G. & Reznikov, L.M. (1993). Dynamic Vibration Absorbers: Theory and Technical
Applications, Wiley, London.
Preumont, A. (2002). Vibration Control of Active Structures: An Introduction, Kluwer, Dordrecht,
2002.
Rao, S.S. (1995). Mechanical Vibrations, Addison-Wesley, NY.
Sira-Ramirez, H. & Agrawal, S.K. (2004). Differentially Flat Systems, Marcel Dekker, NY.
Sira-Ramirez, H., Beltran-Carbajal, F. & Blanco-Ortega, A. (2008). A Generalized Proportional
Integral Output Feedback Controller for the Robust Perturbation Rejection in a
Mechanical System, e-STA, Vol. 5, No. 4, pp. 24-32.
Soderstrom, T. & Stoica, P. (1989). System Identication, Prentice-Hall, NY.
Sun, J.Q., Jolly, M.R., & Norris, M.A. (1995). Passive, adaptive and active tuned vibration
absorbers

U a survey. In: Transaction of the ASME, 50th anniversary of the design
engineering division, Vol. 117, pp. 234
U42.
Taniguchi, T., Der Kiureghian, A. & Melkumyan, M. (2008). Effect of tuned mass damper on
displacement demand of base-isolated structures, Engineering Structures, Vol. 30, pp.
3478-3488.
Weber, B. & Feltrin, G. (2010). Assessment of long-term behavior of tuned mass dampers by
system identication. Engineering Structures, Vol. 32, pp. 3670-3682.
Wright, R.I. & Jidner, M.R.F. (2004). Vibration Absorbers: A Review of Applications in Interior
Noise Control of Propeller Aircraft, Journal of Vibration and Control, Vol. 10, pp. 1221-
1237.
Yang, Y., Muoa, J., & Altintas, Y. (2010). Optimization of multiple tuned mass dampers to
suppress machine tool chatter, International Journal of Machine Tools & Manufacture,
Vol. 50, pp. 834-842.
3
Seismic Response Reduction of Eccentric
Structures Using Liquid Dampers
Linsheng Huo and Hongnan Li
Dalian University of Technology
China
1. Introduction
The installation of vibration absorbers on tall buildings or other flexible structures can be a
successful method for reducing the effects of dynamic excitations, such as wind or earthquake,
which may exceed either serviceability or safety criteria. Tuned liquid column damper (TLCD)
is an effective passive control device by the motion of liquid in a column container. The
potential advantages of liquid vibration absorbers include: low manufacturing and installation
costs; the ability of the absorbers to be incorporated during the design stage of a structure, or
to be retrofitted to serve a remedial role; relatively low maintenance requirements; and the
availability of the liquid to be used for emergency purposes, or for the everyday function of
the structure if fresh water is used (Hitchcock et al., 1997a, 1997b).
A TLCD is a U-shaped tube of uniform rectangular or circle cross-section, containing liquid.
Vibration energy is transferred from the structure to the TLCD liquid through the motion of
the rigid container exciting the TLCD liquid. And the vibration of a structure is suppressed
by a TLCD through the gravitational restoring force acting on the displaced TLCD liquid
and the energy is dissipated by the viscous interaction between the liquid and the rigid
container, as well as liquid head loss due to orifices installed inside the TLCD container.
Analytical and experimental researches on this type of vibration reduction approach has
been conducted, in which viscous interaction between a liquid and solid boundary has been
investigated and used to control vibration (Sakai et al., 1989; Qu et al., 1993). Their
experiments, defining the relationship between the coefficient of liquid head loss (as well as
its dependence on the orifice opening ratio) and the liquid damping, confirms the validity of
their proposed equation of motion in describing liquid column relative motion under
moderate excitation. A variation of TLCD, called a liquid column vibration absorber (LCVA)
has also been investigated, which has different cross sectional areas in its vertical and
horizontal sections depending on performance requirements (Gao and Kwok, 1997; Yan et
al., 1998; Chang and Hsu, 1998; Chang 1999). Yan et al. presented the adjustable frequency
tuned liquid column damper by adding springs to the TLCD system, which modified the
frequency of TLCD and expended its application ranges (Yan and Li, 1999).
Multiple tuned mass damper (MTMD) which consists of a number of tuned mass damper
whose natural frequencies are distributed over a certain range around the fundamental
frequency of the structure has been proposed and investigated (Kareem and Kline, 1995).
The results showed that an optimized MTMD can be more efficient than a single optimized
TMD and the sensitivity of a MTMD to the tuning ratio is diminished. A multiple tuned


48
liquid damper (MTLD) system is investigated by Fujino and Sun(Fujino and Sun, 1993).
They found that in situations involving small amplitude liquid motion the MTLD has
similar characteristics to that of a MTMD including more effectiveness and less sensitivity to
the frequency ratio. However, in a large liquid motion case, the MTLD is not much more
effective than a single optimized TLD and a MTLD has almost the same effectiveness as a
single TLD when breaking waves occur. Gao et al analyzed the characteristics of multiple
liquid column dampers (both U-shaped and V-shaped types) (Gao et al., 1999). It was found
that the frequency of range and the coefficient of liquid head loss have significant effects on
the performance of a MTLCD; increasing the number of TLCD can enhance the efficiency of
MTLCD, but no further significant enhancement is observed when the number of TLCD is
over five. It was also confirmed that the sensitivity of an optimized MTLCD to its central
frequency ratio is not much less than that of an optimized single TLCD to its frequency
ratio, and an optimized MTLCD is even more sensitive to the coefficient of head loss.
2. Circular Tuned Liquid Dampers
Circular Tuned Liquid Column Dampers (CTLCD) is a type of damper that can control the
torsional response of structures (Jiang and Tang, 2001). The results of free vibration and
forced vibration experiments showed that it is effective to control structural torsional
response (Hochrainer et al., 2000), but how to determine the parameters of CTLCD to
effectively reduce torsionally coupled vibration is still necessary to be further investigated.
In this section, the optimal parameters of CLTCD for vibration control of structures are
presented based on the stochastic vibration theory.
2.1 Equation of motion for control system
The configuration of CTLCD is shown in Fig.1. Through Lagrange principle, the equation of
motion for CTLCD excited by seismic can be derived as
( )
( )
2
1
2 2 2 2
2
g
A H R h A h h Agh A R u u

+ + + = +

(1)
where h is the relative displacement of liquid in CTLCD; means the density of liquid; H
denotes the height of liquid in the vertical column of container when the liquid is quiescent;
A expresses the cross-sectional area of CTLCD; g is the gravity acceleration; R represents the
radius of horizontal circular column; is the head loss coefficient; u
denotes the torsional

acceleration of structure;
g
u

is the torsional acceleration of ground motion.
Because the damping in the above equation is nonlinear, equivalently linearize it and the
equation can be re-written as
( )
T Teq T T g
m h c h k h m R u u

+ + = +

(2)
where
T ee
m AL =
is the mass of liquid in CTLCD; 2 2
ee
L H R = + denotes the total length of
liquid in the column;
2
Teq T T T
c m =
is the equivalent damping of CTLCD;
2 /
T ee
g L =
is
the natural circular frequency of CTLCD;
2
T
h
ee
gL
=

is equivalent linear damping ratio

Seismic Response Reduction of Eccentric Structures Using Liquid Dampers

49
(Wang, 1997);
h

means the standard deviation of the liquid velocity; 2
T
k Ag = is the
stiffness of liquid in vibration; 2 /
ee
R L = is the configuration coefficient of CTLCD.

A
h
h
R
x
o
y
z
H
Orifice

Fig. 1. Configuration of Circular TLCD
For a single-story offshore platform, the equation of torsional motion installed CTLCD can
be written as

g
J u c u k u J u F

+ + = +
(3)
where J
is the inertia moment of platform to vertical axis together with additional inertial
moment of sea fluid; c
denotes the summation of damping of platform and additional

damping caused by sea fluid; k
expresses the stiffness of platform; u
and u
are velocity
and displacement of platform, respectively; F
is the control force of CTLCD to offshore

platform, given by

( )
T g
F m R Ru Ru h

= + +

(4)
Combining equation (1) to (4) yields:
2
2 2
2 2
1 / 2 0
0 1
/
/ / 0 2 /
0 /
g
T T
T
R u u u
u
h R h h R R R
R

+ +

+ + =

(5)
where
2
T
m R
J
= denotes inertia moment ratio. Let ( )

i t
g
u t e
= , then

( )
( )
i t
h
u H
e
h H

=

(6)


50
where ( ) H
and ( )
h
H are transfer functions in the frequency domain. Substituting
equation (6) into equation (5) leads to

2 2 2
2 2 2 2 2 2
(1 ) 2 / 1
/
/ / 2 / /
s
h
T T T
i R H
H R
R R i R R

+ + + +
=

+ +

(7)
From the above equation, the transfer function of structural torsional response can be
expressed by

2 2 2 2
2 2 2 2 2 2 4
(1 ) 2(1 ) (1 )
( )
(1 ) 2 2
T T T
T T T
i
H
i i

+ + +

=

+ + + + +

(8)
Then, the torsional response variance of structure installed CTLCD can be obtained as

2
2
( )
g
u u
H S d

=

(9)
If the ground motion is assumed to be a Gauss white noise random process with an intensity
of
0
S and define the frequency ratio /
T
= , the value of
2
u
can be calculated by
2 0
3
4 4 2 3 2 2 2 2 2
1 1 1
2 4 2 3 2 2
1 1 1
2
2(1 ) 2 (1 ) 2 (1 ) (2 ) 2 2 (1 )
2(1 ) 2 4 2 2
u
T T T
T T T T
S
A B C
A B D

=
+ + + + + + + + +
+ + + + +
(10)

where
2 2
1
4 (1 )
T
A = + + ,
2 2 2
1
(2 1)(1 ) 2
T
B

= + + + ,
2 2 4 2
1
4 (1 )
T
C = + + ,
2 2
1
4 D

= +
2.2 Optimal parameters of circular tuned liquid column dampers
The optimal parameters of CTLCD should make the displacement variance of offshore
platform
2
u
minimum, so the optimal parameters of CTLCD can be obtained according to

the following condition

2
0
u
T

2
0
u
(11)
Neglecting the damping ratio of offshore platform
and solving above equation, the

optimal damping ratio
opt
T
and frequency ratio
opt
for CTLCD can be formulized as

2 2
2
5
(1 )
1
4
3
2
(1 )(1 )
2
opt
T

+
=
+ +

2
3
1
2
1
opt

+
=
+
(12)


51
Fig. 2 shows the optimal damping ratio
opt
T
and optimal frequency ratio
opt
of CTLCD as
a function of inertia moment ratio ranging between 0 to 5% for =0.2, 0.4, 0.6 and 0.8. It
can be seen that as the value of increases the optimal damping ratio
opt
T
increases and
the optimal frequency ratio
opt
decreases. For a given value of , the optimal damping
ratio
opt
T
increases and the optimal frequency ratio decreases with the rise of . It can also
be seen that the value of
opt
is always near 1 for different values of and in Fig.2. If let
1 = and solve
2
0
T
, the optimal damping ratio of CTLCD

opt
T
is obtained as

2
3
1 ( 1 ) ( )
2 (1 )
2 2 2
opt
T
1
+ + + +
=
+
(13)

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Inertia moment ratio (%)
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
T
h
e

o
p
t
i
m
a
l

d
a
m
p
i
n
g

r
a
t
i
o

T
o
p
t
=0.6
=0.4
=0.8
=0.2

(a) The optimal damping ratio with inertia moment
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Inertia moment ratio%
0.95
0.955
0.96
0.965
0.97
0.975
0.98
0.985
0.99
0.995
1
T
h
e

o
p
t
i
m
a
l

f
r
e
q
u
e
n
c
y

r
a
t
i
o

o
p
t
=0.2
=0.4
=0.6
=0.8

(b) The optimal frequency ratio with inertia moment
Fig. 2. The optimal parameters of CTLCD with inertia moment ratio


52
The optimal parameters of CTLCD cannot be expressed with formulas when considering the
damping of offshore platform for the complexity of equation (10), so we can only get
numerical results for different values of structural damping, as shown in Table 1. Table 1
shows that for different damping of platform system, the optimal damping ratio of CTLCD
increases and the optimal frequency ratio decreases with the rise of , which is the same as
Fig. 2. Table 1 also suggests the damping of platform has little effect on the optimal
parameters of CTLCD, especially on the optimal damping ratio
opt
T
.

0
= 1%
= 2%
= 5%
=
opt

opt
T

opt

opt
T

opt

opt
T

opt

opt
T

0.5% =
0.9951 0.0282 0.9935 0.0283 0.9915 0.0283 0.9832 0.0283
1% =
0.9903 0.0398 0.9881 0.0398 0.9856 0.0398 0.9755 0.0398
1.5% =
0.9855 0.0487 0.9829 0.0487 0.9799 0.0487 0.9687 0.0487
2% =
0.9808 0.0561 0.9778 0.0561 0.9745 0.0561 0.9622 0.0561
5% =
0.9533 0.0876 0.9490 0.0877 0.9442 0.0877 0.9278 0.0877
Table 1. The optimal parameters of CLTCD ( 0.8 = )
2.3 Analysis of structural torsional response control using CTLCD
The objective of dampers installed in the offshore platform is to increase the damping of the
structural system and reduce the response of structure. To analyze the effects of different
system parameters on the torsional response of structure, the damping of a platform
structure with CTLCD is expressed by equivalent damping ratio
e
(Wang, 1997):

0
e
3 2
= (14)
The relationships between
e
and different parameters of control system are shown in Fig. 3
to Fig. 7.
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
The damping ratio of CTLCD
T
0.01
0.015
0.02
0.025
0.03
0.035
T
h
e

s
t
r
u
c
t
u
r
a
l

e
q
u
i
v
a
l
e
n
t

d
a
m
p
i
n
g

r
a
t
i
o

e
=0.01
=1
=0.8
=0.02
=0.01
=0.015
=0.005

Fig. 3. The structural equivalent damping ratio with the damping ratio of CTLCD


53
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
T
0
0.005
0.01
0.015
0.02
0.025
0.03
T
h
e

s
t
r
u
c
t
u
r
a
l

e
q
u
i
v
a
l
e
n
t

d
a
m
p
i
n
g

r
a
t
i
o

e
=0.01
=0.8
=0.01
=0.6
=0.8
=0.9
=1.0


0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
T
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
T
h
e

s
t
r
u
c
t
u
r
a
l

e
q
u
i
v
a
l
e
n
t

d
a
m
p
i
n
g

r
a
t
i
o

e
=0.01
=1
=0.8
=0.005
=0.01
=0.03
=0.05


0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Inertia moment ratio
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
0.05
0.055
T
h
e

s
t
r
u
c
t
u
r
a
l

e
q
u
i
v
a
l
e
n
t

d
a
m
p
i
n
g

r
a
t
i
o

e
=0.01
=0.1
=1
=0.5
=0.7
=0.6
=0.8
=0.9

Fig. 6. The structural equivalent damping ratio with the inertia moment ratio


54
Fig. 3 shows the equivalent damping ratio of a platform structure
e
as a function of the
damping ratio of CTLCD for =0.005, 0.01, 0.015, 0.02. It is seen from the figure that the
equivalent damping ratio
e
increases rapidly with the increase of
T
initially, whereas it
decreases if the damping ratio of CTLCD
T
is greater than a certain value.
Fig. 4 shows the equivalent damping ratio of a platform structure
e
as a function of the
damping ratio of CTLCD for =0.6, 0.8, 0.9, 1.0. It is seen from the figure that the value of
e
increases with the rise of frequency ratio .
Fig. 5 shows the equivalent damping ratio of structure
e
as a function of the damping ratio
of CTLCD for
=0.005, 0.01, 0.03 and 0.05. It is seen from the figure that as the rise of the
damping ratio of structure
, the equivalent damping ratio

e
increases.
e
as a function of for =0.5, 0.6,
0.7, 0.8 and 0.9. It can be seen from the figure that the damping ratio of structure
e

increases with initially. Whereas, the curve of
e
with will be gentle when the value of
is greater than a certain value. It can also be concluded from the figure that the damping
ratio of structure
e
increases with the rise of configuration coefficient .

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Frequency ratio
0.005
0.01
0.015
0.02
0.025
0.03
0.035
T
h
e

s
t
r
u
c
t
u
r
a
l

e
q
u
i
v
a
l
e
n
t

d
a
m
p
i
n
g

r
a
t
i
o

e
=0.01
=0.1
=0.8
=0.005
=0.015
=0.01
=0.02

Fig. 7. The structural equivalent damping ratio with the frequency ratio
e
as a function of frequency ratio
between CTLCD and structure. It is seen from the figure that the value of
e
will be
maximum at the condition of 1 = . So, the value of can be set to approximate 1 in the
engineering application to get the best control performance.
2.4 Structural torsionally coupled response control using CTLCD
The torsional response of structure is usually coupled with translational response in
engineering, so it is necessary to consider torsionally coupled response for vibration control
of eccentric platform structure. In this paper, a single-story structure only eccentric in x
direction is taken as an example, which means that the displacement in y direction is
coupled with the torsional response of platform. The equation of torsionally coupled motion
for the eccentric platform installed CTLCD can be written as


55

11 12
2 2
21 22
0 0
0 0
y y s gy s s y y y y
y s g
s s
K K e u m m u u u F C C
C C K e K u
u u u F m r m r

+ + = +

(15)
The above equation can be simplified as

+ + = +
s s s s g
M u C u K u M u F
(16)
where
s
m means the mass of platform together with additional mass of sea fluid;
s
e is
eccentric distance;
y
u ,
gy
u and
y
K are the displacement, ground acceleration and stiffness
of offshore platform in y direction, respectively; The control force F is calculated by

( )
( )
y T y gy
T g
F m u u
F m R Ru Ru h

= +

= + +

(17)
It is assumed that the damping matrix in Equation (16) is directly proportional to the
stiffness matrix, that is

a =
s s
C K
(18)
where the proportionality constant a has units of second. The proportionality constant a was
chosen such that the uncoupled lateral mode of vibration has damping equal to 2% of
critical damping. This was to account for the nominal elastic energy dissipation that occurs
in any real structure (Bugeja et al., 1997). The critical damping coefficient c
c
for a single
degree-of-freedom (SDOF) system is given by

2
c s y
c m =
(19)
where
/
y y s
K m =
is natural frequency of the uncoupled lateral mode. From the equation
(18) and (19), the constant a is determined by

0.02 2
y
s
s
y
K
m
m
a
K
=
(20)
Combining the Equation (2) and (15), the equation of motion for torsionally coupled system
can be written as

2 2
2
2
2
2 2 2
2 2
2
2
2
2 2
2
0
1 0 0
0 1 0
0 0 0 2
0
1 0
0 0 1
0
0 0
y y s
y y
y s
T T
y y s
y
y s
T
a a e
u u
a e
R R
u a u
r r r
h h
R
e
u
e
R
u
r r
h

+

+ + +

+

= +

gy
g
u
u
R

(21)


56
where /
T s
m m = is a ratio between the mass of CTLCD and the mass of structure;
2
/( )
T
k m r

= denotes the natural frequency of the uncoupled torsional mode. The
following assumptions are made in this paper:
gy
u and
g
u

are two unrelated Gauss white
noise random processes with intensities of
1
S and
2
S , respectively;
yy
H ,
y
H

,
y
H
and
H
are transfer functions from

gy
u to
y
u ,
g
u

to
y
u ,
gy
u to u
and
g
u

to u
,
respectively. Then, the displacement variance of structure can be obtained by

2 2
2 2
1 2
2
2
2 2
1 2
yy y
y
u yy u y
u y u
S H d S H d
S H d S H d

= =
= =

(22)
where
yy
u
and
y
u

are displacement variances in y direction caused by the ground
motion in y direction and direction, respectively;
y
u
and
u
are displacement
variances in direction caused by the ground motion in y direction and direction,
respectively. So, the equivalent damping ratios of structure are given as

1
3 2
2
yy
eyy
y u
S

=
2
2
3 2
2
y
ey
y u
S r

=
1
3 2 2
2
y
e y
y u
S
r

=
2
3 2
2
e
y u
S

= (23)
where
eyy
and
ey
are equivalent damping ratios in y direction caused by the ground
motions in y direction and direction, respectively;
e y
and
e
are equivalent
damping ratios in direction caused by the ground motions in y direction and
direction, respectively. Then, the total equivalent damping ratio
ey
in y direction and
e

in direction can be defined as

ey eyy ey
e e y e

= +
= +
(24)
Define /
x
= as the frequency ratio between the uncoupled torsional mode and
uncoupled translational mode and
1
as the first frequency of torsionally coupled structure.
The relationships of equivalent damping ratio
ey
and
e
with parameters of control
system are shown in Fig. 8 to Fig. 11.
Fig. 8 shows the equivalent damping ratio
ey
and
e
as functions of frequency ratio
1
/
T
for mass ratio =0.005, 0.01, 0.015 and 0.02. It is seen from the figure that the
values of
ey
and
e
are maximum when the value of frequency ratio
1
/
T
is
approximate 1. The Fig. 8 also suggests that damping ratio
ey
and
e
increase with the
rise of mass ratio .
Fig. 9 shows equivalent damping ratio
ey
and
e
as functions of mass ratio for
configuration coefficient =0.5, 0.6, 0.7 and 0.8. It is seen from the figure that the values of
ey
and
e
increase initially and approach constants finally with the rise of mass ratio .


57
It can also be concluded that the values of
ey
and
e
both increase with the rise of
configuration coefficient .

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Frequency ratio
/
1
0.018
0.02
0.022
0.024
0.026
0.028
0.03
0.032
0.034
0.036
0.038
0.04
E
q
u
i
v
a
l
e
n
t

d
a
m
p
i
n
g

r
a
t
i
o

i
n

y

d
i
r
e
c
t
i
o
n

e
y
=0.02
=0.01
=0.015
=0.005
=0.8
e
s
/r=0.5
=0.8
T
=0.02

(a) Equivalent damping ratio in y direction

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Frequency ratio
/
1
0.01
0.012
0.014
0.016
0.018
0.02
0.022
0.024
E
q
u
i
v
a
l
e
n
t

d
a
m
p
i
n
g

r
a
t
i
o

i
n

d
i
r
e
c
t
i
o
n

=0.02
=0.01
=0.015
=0.005
=0.8
e
s
/r=0.5
=0.8
T
=0.02

(b) Equivalnet damping ratio in direction

Fig. 8. Equivalent damping ratio of structure with frequency ratio
1
/
T


58
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
Mass ratio
0
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016
0.018
0.02
E
q
u
i
v
a
l
e
n
t

d
a
m
p
i
n
g

r
a
t
i
o

i
n

y

d
i
r
e
c
t
i
o
n

e
y
=0.8
e
s
/r=0.5
/
1
=1
T
=0.02
=0.5
=0.6
=0.8
=0.7


0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
Mass ratio
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016
E
q
u
i
v
a
l
e
n
t

d
a
m
p
i
n
g

r
a
t
i
o

i
n

d
i
r
e
c
t
i
o
n

=0.5
=0.6
=0.8
=0.7
=0.8
e
s
/r=0.5
/
1
=1
T
=0.02

(b)Equivalnet damping ratio in direction

Fig. 9. Equivalent damping ratio of structure with mass ratio


59
0 0.05 0.1 0.15 0.2
Damping ratio of CTLCD
T
0.0196
0.01965
0.0197
0.01975
0.0198
0.01985
0.0199
0.01995
0.02
0.02005
0.0201
0.02015
E
q
u
i
v
a
l
e
n
t

d
a
m
p
i
n
g

r
a
t
i
o

i
n

y

d
i
r
e
c
t
i
o
n

e
y
=0.5
=0.6
=0.8
=0.7
=0.8
e
s
/r=0.5
/
1
=1
=0.01


0 0.05 0.1 0.15 0.2
Damping ratio of CTLCD
T
0.01069
0.0107
0.01071
0.01072
0.01073
0.01074
0.01075
0.01076
0.01077
0.01078
0.01079
E
q
u
i
v
a
l
e
n
t

d
a
m
p
i
n
g

r
a
t
i
o

i
n

d
i
r
e
c
t
i
o
n

=0.8
e
s
/r=0.5
/
1
=1
=0.01
=0.5
=0.6
=0.8
=0.7


Fig. 10. Equivalent damping ratio of structure with damping ratio
T

Fig. 10 shows equivalent damping ratio
ey
and
e
as functions of damping ratio
T
for
=0.5, 0.6, 0.7 and 0.8. It is seen from the figure that the values of
ey
and
e
rapidly
increase initially with the rise of
T
; whereas, after a certain value of
T
,
ey
will decrease
to a constant and
e
decrease first, then increase gradually.


60
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Frequency ratio
0
0.02
0.04
0.06
0.08
0.1
E
q
u
i
v
a
l
e
n
t

d
a
m
p
i
n
g

r
a
t
i
o

i
n

y

d
i
r
e
c
t
i
o
n

e
y
e
s
/r=0.4
e
s
/r=0.6
e
s
/r=0.8
e
s
/r=1.0
=0.8
T
=0.2
/
1
=1
=0.01

0 0.2 0.4 0.6 0.8 1 1.2 1.4
Frequency ratio
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
E
q
u
i
v
a
l
e
n
t

d
a
m
p
i
n
g

r
a
t
i
o

i
n

d
i
r
e
c
t
i
o
n

=0.8
T
=0.2
/
1
=1
=0.01
e
s
/r=0.4
e
s
/r=0.6
e
s
/r=0.8
e
s
/r=1.0

Fig. 11. Equivalent damping ratio of structure with frequency ratio
Fig 11 shows equivalent damping ratio
ey
and
e
as functions of frequency ratio for
/
s
e r =0.4, 0.6 0.8 and 1.0. It is seen from the figure that the values of
ey
and
e
are
approximate zero for the structure with near to /
s
e r ; for the structure /
s
e r < , the
values of
ey
and
e
decrease with the rise of frequency ratio and increase with the rise
of /
s
e r ; for the structure with /
s
e r > , the values of
ey
and
e
increase with the rise of
frequency ratio and decrease with the rise of /
s
e r .
3. Torsionally coupled vibration control of eccentric buildings
The earthquake is essentially multi-dimensional and so is the structural response excited by
earthquake, which will result in the torsionally coupled vibration that cannot be neglected.
So, the torsional response for structure is very important (Li and Wang, 1992). Previously, Li
et al. presented the method of reducing torsionally coupled response by installing TLCDs in
structural orthogonal directions (Huo and Li, 2001). Circular Tuned Liquid Column


61
Dampers (CTLCD) is a type of control device sensitive to torsional response. The results of
free vibration and forced vibration experiments showed that it is effective to control
structural torsional response (Liang, 1996; Hochrainer, 2000). However, how to determine
the parameters of CTLCD to effectively reduce torsionally coupled vibration is still
necessary to be further investigated.
3.1 Equation of motion for control system
The configuration of TLCD is shown in Fig. 12(a). According to the Lagrange theory, the
equation of motion for TLCD excited by seismic can be derived as
( )
( )
1
2 2
2
g
A H B h A h h Agh AB u u + + + = +

(25)
where h is the relative displacement of liquid in TLCD; means the density of liquid; H
expresses the height of liquid in the container when the liquid is quiescent; A denotes the
cross-sectional area of TLCD; g is the acceleration of gravity; B represents the length of
horizontal liquid column; is the head loss coefficient; u and
g
u mean the acceleration of
structure and ground motion, respectively

(a) (b)
Fig. 12. Configuration of Liquid Column Dampers
The shape of CTCD is shown in Fig. 12(b). In the same way, the equation of motion for
CTLCD is derived as
( )
( )
2
1
2 2 2 2
2
g
A H R h A h h Agh A R u u

+ + + = +

(26)
Two TLCDs are set in the longitudinal direction and transverse direction of n-story building,
respectively, and a CTLCD is installed in the center of mass, as shown in Fig.13. The
equation of motion of system excited by multi-dimensional seismic inputs can be written as
[ ]{ } [ ]{ } [ ]{ } [ ][ ]{ }
{ }
s s s s s g T
M u C u K u M E u F + + = + (27)
Where, [ ]
s
M , [ ]
s
C and [ ]
s
K are the mass, damping and stiffness matrices of the system
with dimension of 3n3n, respectively. { } u means hte displacement vector of the strucutre,
{ }
{ }
1 1 1
T
x xn y yn n
u u u u u u u

= " " " ; [Es] is the influence matrix of the ground


62
excitation;
{ } { }
g xg yg g
u u u u
= is the three-dimensional siesmic inputs;

{ }
0 0 0
T x y
F F F F
= " " " is the three-dimensional control vector, where

1 2
1 2
1 2 1 2
1
( ) ( )( )
( ) ( )( )
( )( ) ( )( )
x Ttot xn xg x Tx x Tx y Ty y n g
y Ttot yn yg y Ty y Tx x Ty x n g
Tx y Ty y xn xg Tx x Ty x yn yg
x Tx y x y T
F m u u m h m l m l u u
F m u u m h m l m l u u
F m l m l u u m l m l u u
m l h m

= + + + +
= + + +
= + + + +
+
2 2 2
2 1 2
( )( )
y x y Tx Ty T n g T
l h m r m r m R u u m Rh

+ + +

(28)
where (
1x
l ,
1y
l ) means the location of the TLCD in x direction; (
2x
l ,
2y
l ) means the
location of the TLCD in y direction;
2 2 2
1 1 1 x y
r l l = + ;
2 2 2
2 2 2 x y
r l l = + .
Combining Equation (1) to (3), the equation of motion for the control system can be written as
+ + =
g
Mx Cx Kx MEu (29)
where M, C and K are the mass, damping and stiffness matrices of the combined and
damper system. Although the damping of the structure is assumed to be classical, the
combined structure and damper system represented by the above equation will be non-
classically damped. To analyze a non-classical damped system, it is convenient to work with
the system of first order state equations
= +
g
Z AZ Bu
(30)
where

=

x
Z
x

1 1

=

0 I
A
M K M C

=

0
B
E
(31)

Fig. 13. An eccentric structure with liquid Dampers


63
3.2 Dynamical characteristics of the structure
The structure analyzed in this paper is an 8-story moment-resisting steel frame with a plan
irregularity and a height of 36m created in this study and shown in Figure 14 and Figure
15(Kim, 2002). The structure has 208 members, 99 nodes, and 594 DOFs prior to applying
boundary conditions, rigid diaphragm constrains, and the dynamic condensation. Applying
boundary conditions and rigid diaphragm constraints results in 288 DOFs. They are further
reduced 24 DOFs by the Guyan reduction of vertical DOFs and the rotational DOFs about
two horizontal axes.
The static loading on the building consists of uniformly distributed floor dead and live load
of 4.78 Kpa and 3.35 Kpa, respectively. A total lateral force (base force) of 963 KN is obtained
and distributed over the structure using the equivalent linear static load approach. Each
floor shear force is distributed to the nodes in that floor in proportion to nodal masses.

Fig. 14. Plan of the structure

Fig. 15. FEM figure of the structure


64

(a) Mode 1 (b) Mode 2 (c)Mode 3
Fig. 16. The first three modes of the structure
Because of plan irregularity substantially more translational and torsional coupling effect is
expected in this example. Figure 16 shows the first three modes of vibrations: (a) mode 1
with a frequency of 0.57 Hz, (b) mode 2 with a frequency of 0.72 Hz, (c) mode 3 with a
frequency of 0.75 Hz.
3.3 Optimization of the damper parameters
To reduce the torsionally coupled vibration of the 8-story eccentric buildings, two TLCDs
are respectively installed on the top story of the structure along x and y direction and one
CTLCD on the mass center of the top story. Hence, there are many parameters of the control
system to be optimized. The focus of the paper is how to optimize the parameters of liquid
dampers to effectively control the dynamical responses of structures. Genetic Algorithm
(GA) provides a general framework for the optimization of complicated systems, which is
independent of specific areas and robust for the types of problems. In this section,
parameters of liquid dampers are optimized by use of GA.
The following form of performance function is used, primarily because it is easy to evaluate
the responses of structures

2 2 2
,controlled
1
2
2 2 2
,uncontrolled
1
( )
1
( )
n
x y i
i
n
x y i
i
d d d
f
d d d
=
=
+ +
=
+ +

1, 2, 3, , i n = "
(32)
where,
x
d ,
y
d and d
are the drifts of a story in x, y and torsion direction.

Both, the stochastic model and the design response spectra have been used to define the
base input motion. For the stochastic model, the ground motions in x and y directions are
descried by two identical but uncorrelated zero-mean stationary processes with power
spectral density function ( )
l
of the Kanai-Tajimi form:


65

2
2
0
2
2 2
2
1 4
( )
1 4
g
l
g g
S

+

=

+

(33)
The parameters of this function are: 5
g
= , 0.5 = ,
0
0.01 S = .
As mentioned before, two TLCDs and one CTLCD are installed at the top of the building.
The objective of the study is to design the optimum parameters of these dampers that would
maximize the performance function stated earlier. The possible ranges for the design
parameters are fixed as follows:
1. Mass ratio, : The mass ratio is defined as the ratio of the damper mass to the total
building mass. It is assumed that each damper ratio can vary in the range of 0.1 percent
to 1 percent of the building mass. Thus the maximum mass of the damper system
consisting of three dampers could be as high as 3 percent of the building mass.
2. Frequency tuning ratio, f: The frequency ratio for each damper is defined as the ratio its
own natural frequency to the fundamental frequency of the building structure. Here it
is assumed that this ratio could vary between 0-1.5.
3. Damping ratio, d: This is a ratio of the damping coefficient to its critical value. It is
assumed that this ratio can vary in the range of 0-10 percent.
4. Damper positions from the mass center, lx in x axis and ly: in y axis: It is assumed that lx
can vary between 8 and 5 meters and ly can vary between 4 and 3 meters
The optimization process starts with a population of these individuals. For the problem at
hand, 30 individuals were selected to form the population. The probability of crossover and
mutation are 0.95 and 0.05, respectively. The process of iteration is determined to be 300 steps.
The final optimum parameters for the two optimum design criteria are given in Table 2.

Performance Criteria i (f1=0.47769)
TLCD in x direction TLCD in y direction CTLCD

0.008519 0.0095655 0.0014362
f
r
1.2334 0.96607 1.1137
d
r
0.053803 0.061988 0.052886
l
x
-7.38 0.45567
l
y
-6.479 -2.2431
Table 2. The optimal parameters of liquid dampers
3.4 Seismic analysis in time domain
The parameters of liquid dampers on the 8-story building structure have been optimized in
the previous section and the results are listed in the Table 2. The control results of liquid
dampers on the building are analyzed in time domains in this section. The El Centro, Tianjin
and Qianan earthquake records are selected to input to the structure as excitations, which
represent different site conditions.
The structural response without liquid dampers subjected to earthquake in x, y and
directions are expressed with x
0
, y
0
and
0
, respectively. Also, the response with liquid


66
dampers subjected to earthquake in x, y and directions are expressed with x, y and ,
respectively. The response reduction ratio of the structure is defined as

0
0
100%
x x
J
x
=
(34)
The maximum displacements of the structure and response reduction ratios are computed
for three earthquake records and the results listed from Table 3 to Table 5. It can be seen

Story
Number
x
0
(cm)
x
(cm)
J
(%)
y
0
(cm)
y
(cm)
J
(%)
0
(10
-4
Rad)

(10
-4
Rad)
J
(%)
1 0.57 0.53 6.14 1.18 0.99 15.94 1.71 1.56 8.77
2 1.11 1.03 7.24 2.28 1.94 15.12 3.30 3.04 7.88
3 1.77 1.58 10.71 3.43 2.96 13.73 5.02 4.61 8.17
4 2.38 2.04 14.20 4.30 3.76 12.59 6.42 5.87 8.57
5 2.98 2.54 15.01 4.95 4.39 11.34 7.62 6.90 9.45
6 3.48 3.03 12.77 5.64 4.69 16.76 8.50 7.46 12.24
7 4.03 3.67 8.86 6.58 4.89 25.64 10.05 8.63 14.13
8 4.37 4.06 7.00 7.10 5.47 22.98 10.96 9.63 12.14
Table 3. Maximum displacements of the structure (El Centro)

Story
Number
x
0
(cm)
x
(cm)
J
(%)
y
0
(cm)
y
(cm)
J
(%)
0
(10
-4
Rad)

(10
-4
Rad)
J
(%)
1 2.49 1.85 25.64 2.10 1.83 12.57 4.67 4.22 9.64
2 4.91 3.66 25.49 4.05 3.52 13.03 9.14 8.25 9.74
3 7.71 5.78 25.06 6.30 5.48 13.12 14.21 12.73 10.42
4 10.26 7.69 24.98 8.36 7.23 13.59 18.83 16.75 11.05
5 12.77 9.58 24.97 10.36 8.90 14.08 23.21 20.60 11.25
6 14.79 11.09 24.97 11.98 10.27 14.25 26.55 23.56 11.26
7 16.88 12.66 24.96 14.10 12.13 13.91 30.45 27.05 11.17
8 17.98 13.51 24.89 15.22 13.41 11.90 32.44 29.02 10.54
Table 4. Maximum displacements of the structure (Tianjin)

Story
Number
x
0
(cm)
x
(cm)
J
(%)
y
0
(cm)
y
(cm)
J
(%)
0
(10
-4
Rad)

(10
-4
Rad)
J
(%)
1 0.11 0.10 6.91 0.10 0.091 6.93 0.132 0.12 4.10
2 0.19 0.17 6.57 0.19 0.16 16.00 0.25 0.24 1.25
3 0.23 0.21 6.66 0.28 0.22 24.60 0.39 0.38 2.82
4 0.24 0.21 11.48 0.38 0.28 25.86 0.51 0.50 1.45
5 0.29 0.22 23.54 0.48 0.36 24.89 0.63 0.61 1.40
6 0.34 0.26 24.76 0.56 0.43 23.54 0.72 0.70 2.20
7 0.39 0.34 14.17 0.70 0.54 21.70 0.83 0.79 4.45
8 0.43 0.39 8.15 0.77 0.62 19.47 0.93 0.87 5.60
Table 5. Maximum displacements of the structure (Qianan)


67

Fig. 17. Time history of the displacement on the x direction of top floor (El Centro)

Fig. 18. Time history of the displacement on the y direction of top floor (El Centro)

Fig. 19. Time history of the torsional displacement of top floor (El Centro)


68
from the tables that the responses of the structure in each degree of freedom are reduced
with the installation of liquid dampers. However, the reduction ratios are different for the
different earthquake records.
The displacement time history curves of the top story are shown from Fig. 6 to Fig. 8 and
acceleration time history curves in Fig. 17 to Fig. 19 for El Centro earthquake. It can be seen
from these figures that the structural response are reduced in the whole time history.
4. Conclusion
From the theoretical analysis and seismic disasters, it can be concluded that the seismic
response is not only in translational direction, but also in torsional direction. The torsional
components can aggravate the destroy of structures especially for the eccentric structures.
Hence, the control problem of eccentric structures under earthquakes is very important. This
paper focus on the seismic response control of eccentric structures using tuned liquid
dampers. The control performance of Circular Tuned Liquid Column Dampers (CTLCD) to
torsional response of offshore platform structure excited by ground motions is investigated.
Based on the equation of motion for the CTLCD-structure system, the optimal control
parameters of CTLCD are given through some derivations supposing the ground motion is
stochastic process. The influence of systematic parameters on the equivalent damping ratio
of the structures is analyzed with purely torsional vibration and translational-torsional
coupled vibration, respectively. The results show that Circular Tuned Liquid Column
Dampers (CTLCD) is an effective torsional response control device. An 8-story eccentric
steel building, with two TLCDs on the orthogonal direction and one CTLCD on the mass
center of the top story, is analyzed. The optimal parameters of liquid dampers are optimized
by Genetic Algorithm. The structural response with and without liquid dampers under bi-
directional earthquakes are calculated. The results show that the torsionally coupled
response of structures can be effectively suppressed by liquid dampers with optimal
parameters.
5. Acknowledgment
This work was jointly supported by Natural Science Foundation of China (no. 50708016 and
90815026), Special Project of China Earthquake Administration (no. 200808074) and the 111
Project (no. B08014).
6. References
Bugeja, N.; Thambiratnam, D.P. & Brameld G.H. (1999). The Influence of Stiffness and
Strength Eccentricities on the Inelastic Earthquake Response of Asymmetric
Structures, Engineering Structures, Vol. 21,No.9, pp.856863
Chang, C. C. & Hsu, C. T.(1998). Control Performance of Liquid Column Vibration
Absorbers. Engineering Structures. Vol.20, No.7, pp.580-586
Chang, C. C.(1999). Mass Dampers and Their Optimal Design for Building Vibration
Control. Engineering Structures, Vol.21, No.5, pp.454-463


69
Fujina, Y. & Sun, L. M.(1993). Vibration Control by Multiple Tuned Liquid
Dampers(MTLDs). J. of Structural Engineering, ASCE, Vol.119, No.12,
pp.3482-3502
Gao H. & Kwok K. C. S.(1997). Optimization of Tuned Liquid Column Dampers. Engineering
Structures, Vol.19, No.6, pp. 476-486
Gao, H.; Kwok, K. S. C. & Samali B.(1999). Characteristics Of Multiple Tuned Liquid
Column Dampers In Suppressing Structural Vibration. Engineering Structures,
Vol.21, No.4, pp.316-331
Hitchcock, P. A.; Kwok, K. C. S. & Watkins R. D. (1997). Characteristics Of Liquid Column
Vibration Absorbers (LCVA)-I. Engineering Structures, Vol.19, No.2, pp.126-134
Hitchcock, P. A.; Kwok, K. C. S. & Watkins R. D. (1997). Characteristics Of Liquid Column
Vibration Absorbers (LCVA)-II. Engineering Structures, Vol.19, No.2,
pp.135-144
Hochrainer, M. J.; Adam, C. & Ziegler, F. (2000). Application of Tuned Liquid Column
Dampers for Passive Structural Control, Proc. of 7th International Congress on Sound
and Vibration (ICSV 7), Garmisch-Partenkirchen, Germany, pp.3107-3114
Huo, L.S. & Li H.N. (2001). Parameter Study of TLCD Control System Excited by Multi-
Dimensional Ground Motions. Earthquake Engineering and Engineering Vibration,
Vol.21, No.4, pp.147-153
Jiang, Y. C. & Tang, J. X. (2001). Torsional Response of the Offshore Platform with TMD,
China Ocean Engineering, Vol.15, No.2, pp.309-314
Kareem, A. & Kline, S.(1995). Performance Of Multiple Mass Dampers Under Random
Loading. J. of Structural Engineering, ASCE, Vol.121, No.2, pp.348-361
Kim, H. (2002). Wavelet-based adaptive control of structures under seismic and wind loads.
Notre Dame: The Ohio State University
Li, H.N. & Wang, S.Y. (1992). Torsionally Coupled Stochastic Response Analysis of Irregular
Buildings Excited by Multiple Dimensional Ground Motions. Journal of Building
Structures, Vol.13, No.6, pp.12-20
Liang, S.G. (1996).Experiment Study of Torsionally Structural Vibration Control Using
Circular Tuned Liquid Column Dampers, Special Structures, Vol.13, No.3,
pp. 33-35
Qu, W. L.; Li, Z. Y. & Li, G. Q. (1993). Experimental Study on U Type Water Tank for Tall
Buildings and High-Rise Structures, China Journal of Building Structures, Vol.14,
No.5, pp.37-43
Sakai, F.; Takaeda, S. & Tamake, T. (1989). Tuned Liquid Column Damper New Type
Device For Suppression Of Building Vibrations. Proc. Int. Conf. On High-rise
Buildings, Nanjing, China, pp.926-931
Wang, Z. M. (1997). Vibration Control of Towering Structures, Shanghai: Press of Tongji
University.
Yan, S. & Li H. N.(1999). Performance of Vibration Control for U Type Water Tank with
Variable Cross Section. Earthquake Engineering and Engineering Vibration, Vol.19,
No.1, pp.197-201


70
Yan, S.; Li, H. N. & Lin, G. (1998). Studies on Control Parameters of Adjustable Tuned
Liquid Column Damper, Earthquake Engineering and Engineering Vibration, Vol18,
No.4, pp. 96-101
4
Active Control of Human-Induced Vibrations
Using a Proof-Mass Actuator
Ivn M. Daz
Universidad de Castilla-La Mancha
Spain
1. Introduction
Advances in structural technologies, including construction materials and design
technologies, have enabled the design of light and slender structures, which have
increased susceptibility to human-induced vibration. This is compounded by the trend
toward open-plan structures with fewer non-structural elements, which have less inherent
damping. Examples of notable vibrations under human-induced excitations have been
reported in floors, footbridges and grandstands, amongst other structures (Bachmann,
1992; Bachmann, 2002; Hanagan et al., 2003a). Such vibrations can cause a serviceability
problem in terms of disturbing the users, but they rarely affect the fatigue behaviour or
safety of structures.
Solutions to overcome human-induced vibration serviceability problems might be: (i)
designing in order to avoid natural frequencies into the habitual pacing rate of walking,
running or dancing, (ii) stiffening the structure in the appropriate direction resulting in
significant design modifications, (iii) increasing the weight of the structure to reduce the
human influence being also necessary a proportional increase of stiffness and (iv) increasing
the damping of the structure by adding vibration absorber devices. The addition of these
devices is usually the easiest way of improving the vibration performance. Traditionally,
passive vibration absorbers, such as tuned mass dampers (TMDs) (Setareh & Hanson, 1992;
Caetano et al., 2010), tuned liquid dampers (Reiterer & Ziegler, 2006) or visco-elastic
dampers, etc., have been employed. However, the performance of passive devices is often of
limited effectiveness if they have to deal with small vibration amplitude (such as those
produced by human loading) or if vibration reduction over several vibration modes is
required since they have to be tuned to a single mode. Semi-active devices, such semi-active
TMDs, have been found to be more robust in case of detuning due to structural changes, but
they exhibit only slightly improved performance over passive TMDs and they still have the
fundamental problem that they are tuned to a single problematic mode (Setareh, 2002;
Occhiuzzi et al., 2008). In these cases, an active vibration control (AVC) system might be
more effective and then, an alternative to traditional passive devices (Hanagan et al., 2003b).
A state-of-the-art review of technologies (passive, semi-active and active) for mitigation of
human-induced vibration can be found in (Nyawako & Reynolds, 2007). Furthermore,
techniques to cancel floor vibrations (especially passive and semi-active techniques) are
reviewed in (Ebrahimpour & Sack, 2005) and the usual adopted solutions to cancel
footbridge vibrations can be found in (FIB, 2005).


72
An AVC system based on direct velocity feedback control (DVFC) with saturation has
been studied analytically and implemented experimentally for the control of human-
induced vibrations via an active mass damper (AMD) (also known as inertial actuator or
proof-mass actuator) on a floor structure (Hanagan & Murray, 1997) and on a footbridge
(Moutinho et al., 2010). This actuator generates inertial forces in the structure without
need for a fixed reference. The velocity output, which is obtained by an integrator circuit
applied to the measured acceleration response, is multiplied by a constant gain and feeds
back to a collocated force actuator. The term collocated means that the actuator and sensor
are located physically at the same point on the structure. The merits of this method are its
robustness to spillover effects due to high-order unmodelled dynamics and that it is
unconditionally stable in the absence of actuator and sensor (integrator circuit) dynamics
(Balas, 1979). Nonetheless, when such dynamics are considered, the stability for high
gains is no longer guaranteed and the system can exhibit limit cycle behaviour, which is
not desirable since it could result in dramatic effects on the system performance and its
components (Daz & Reynolds, 2010a). Then, DVFC with saturation is not such a desirable
solution. Generally, the actuator and sensor dynamics influence the system dynamics and
have to be considered in the design process of the AVC system. If the interaction between
sensor/actuator and structure dynamics is not taken into account, the AVC system might
exhibit poor stability margins, be sensitive to parameter uncertainties and be ineffective.
A control strategy based on a phase-lag compensator applied to the structure acceleration
(Daz & Reynolds, 2010b), which is usually the actual magnitude measured, can alleviated
such problems. This compensator accounts for the interaction between the structure and
the actuator and sensor dynamics in such a way that the closed-loop system shows
desirable properties. Such properties are high damping for the fundamental vibration
mode of the structure and high stability margins. Both properties lead to a closed-loop
system robust with respect to stability and performance (Preumont, 1997). This control
law is completed by: (i) a high-pass filter, applied to the output of the phase-lag
compensator, designed to avoid actuator stroke saturation due to low-frequency
components and (ii) a saturation nonlinearity applied to the control signal to avoid
actuator force overloading at any frequency. This methodology will be referred as to
compensated acceleration feedback control (CAFC) from this point onwards.
This chapter presents the practical implementation of an AMD to cancel excessive vertical
vibrations on an in-service office floor and on an in-service footbridge. The AMD consists of
a commercial electrodynamic inertial actuator controlled via CAFC. The remainder of this
chapter is organised as follows. The general control strategy together with the structure and
actuator dynamic model are described in Section 2. The control design procedure is
described in Section 3. Section 4 deals with the experimental implementation of the AVC
system on an in-service open-plan office floor whereas Section 5 deals with the
implementation on an in-service footbridge. Both sections contain the system dynamic
models, the design of CAFC and results to assess the design. Finally, some conclusions are
given in Section 6.
2. Control strategy and system dynamics
The main components of the general control strategy adopted in this work are shown in Fig.
1. The output of the system is the structural acceleration since this is usually the most
convenient quantity to measure. Because it is rarely possible to measure the system state

Active Control of Human-Induced Vibrations Using a Proof-Mass Actuator

73
and due to simplicity reasons, direct output measurement feedback control might be
preferable rather than state-space feedback in practical problems (Chung & Jin, 1998). In this
Fig., G
A
is the transfer function of the actuator, G is of the structure, C
D
is of the direct
compensator and C
F
is of the feedback compensator. The direct one is merely a phase-lead
compensator (high-pass property) designed to avoid actuator stroke saturation for low-
frequency components. It is notable that its influence on the global stability will be small
since only a local phase-lead is introduced. The feedback one is a phase-lag compensator
designed to increase the closed-loop system stability and to make the system more amenable
to the introduction of significant damping by a closed-loop control. The control law is
completed by a nonlinear element f that is assumed to be a saturation nonlinearity to
account for actuator force overloading.

Fig. 1. General control scheme
2.1 Structure dynamics
If the collocated case between the acceleration (output) and the force (input) is considered
and using the modal analysis approach, the transfer function of the structure dynamics can
be represented as an infinitive sum of second-order systems as follows (Preumont, 1997)
( )
2
2 2
1
2
i
i i i i
s
G s
s s
=
=
+ +
, (1)
where = s j , is the frequency,
i
,
i
and
i
are the inverse of the modal mass,
damping ratio and natural frequency associated to the i-th mode, respectively. For practical
application, N vibration modes are considered in the frequency bandwidth of interest. The
transfer function G (1) is thus approximated by a truncated one as follows
( ) : r t
Reference command ( ) : y t
Acceleration response
( ) : V t
Control voltage ( ) :
c
y t
Compensated acceleration
( ) : F t
Actuator force ( )
0
: V t
Initial control voltage
( ) : p t
Plant disturbance ( ) :
c
f y
Nonlinear element
( ) :
D
C s
Transfer function of the direct compensator
( ) :
A
G s
Transfer function of the AMD
( ) : G s
Transfer function of the structure
( ) :
F
C s
Transfer function of the feedback compensator
( ) p t
( ) F t
( ) 0 r t =

+ ( ) V t
( ) y t
( )
c
y t
+
+
( )
A
G s
( ) G s
( )
0
V t
( )
F
C s
( )
D
C s
( ) ( )
c
f y t


74

( )

=
+ +
2
2 2
1
2
N
i
i
i i i
s
G s
s s
. (2)
2.2 Proof-mass actuator dynamics
The actuator consists of a reaction (moving) mass attached to a current-carrying coil moving
in a magnetic field created by an array of permanent magnets. The linear behaviour of a
proof-mass actuator can be closely described as a linear third-order model (Reynolds et al.,
2009). That is, a low-pass element is added to a linear second-order system in order to
account for the low-pass property exhibited by these actuators. The cut-off frequency of this
element is not always out of the frequency bandwidth of interest since it is approximately 10
Hz (APS). Such a low-pass behaviour might affect importantly the global stability of the
AVC system. Thus, the actuator is proposed to be modelled by

( )
( ) ( )

= =

+ + + + + + + +

2 2
2 2 3 2 2 2
1
2 2 2
A A
A
A A A A A A A A A
K s K s
G s
s s s s s s
(3)
where > 0
A
K , and
A
and
A
are, respectively, the damping ratio and natural frequency
which take into consideration the suspension system and internal damping. The pole at
provides the low-pass property.
3. Controller design
The purpose of this section is to show a procedure to design the compensators C
D
and C
F

(see Fig. 1). The design of C
D
is undertaken in the frequency domain and the design of C
F
is
carried out through the root locus technique. The root locus maps the complex linear system
roots of the closed-loop transfer function for control gains from zero (open-loop) to infinity
(Bolton, 1998). In the design of C
F
, it is assumed that the natural frequency of the actuator
A
(see Eq. (3)) is below the first natural frequency of the structure
1
(see Eq. (2))
(Hanagan, 2005).
3.1 Direct compensator
The transfer function between the inertial mass displacement and input voltage to the
actuator can be considered as follows

( )
( )
=
2
1
A
d
A
G s
G s
m s
, (4)
with m
A
being the mass value of the inertial mass. Fig. 2a shows an example of magnitude of
G
d
. The inertial mass displacement at low frequencies should be limited due to stroke
saturation. A transfer function with the following magnitude is defined
( )
( )

=

< <
d
d
d
G j
G j
, (5)


75
in which d is the maximum allowable stroke for harmonic input per unit voltage and
is
the higher frequency that fulfils ( ) =
d
G j d . A high-pass compensator of the form
( )

+
=
+
, ,
D
s
C s
s
with > 0, (6)
is applied to the initial control voltage
0
V and its output is the filtered input V to the
actuator (see Fig. 1). Fig. 2b shows an example of C
D
. The following error function is defined
( ) ( ) ( ) ( )
( )
=
2
, , , ,
d D d
e G j C j G j , (7)
with ( ) ,
L U
, <
L
, >
U
, and
L
and
U
being, respectively, the lower and
upper value of the frequency range that is considered in the design. The lower frequency
L

must be selected in such a way that the actuator resonance is sufficiently included and the
upper frequency
U
must be chosen so that the structure dynamics that are prone to be
excited are included. Parameters and of the compensator are obtained by minimising the
error function (7)

( )
( ) ( )

+

\
, , , ,
min
,
L U
e , (8)
with [ )
max
0, , [ )
max
0, ,
max
,
max
and
max
and
max
being, respectively, the
maximum considered value of and for the optimisation problem (8). Note that and
are delimited by the low-pass property of the actuator in order to minimise the influence
of C
D
on the global stability properties. By and large, the objective is to fit ( ) ( )
D d
C j G j to
d for < <
L
and not to affect the dynamics for < <
U
(see Fig. 2a). The result is a
high-pass compensator that introduces dynamics mainly in the frequency range < <
L

in such a way that the global stability is not compromised.

Fig. 2. Effect of the high-pass compensator on the actuator dynamics. a) Magnitude of G
d

and C
D
G
d
. b) Magnitude of C
D


76
3.2 Feedback compensator
To illustrate the selection of the form of the compensator C
F
, the root locus map (s-plane) for
four different cases is shown in Fig. 3. A realistic structure is assumed with two significant
vibration modes. The modal mass and damping ratio for both modes are assumed to be 20
tonnes and 0.03, respectively. The four cases are: a) direct acceleration feedback control
(DAFC) considering two vibration modes at 4 and 10 Hz, respectively; b) DVFC for the same
structure as in a); c) DAFC considering two vibration modes at 7 and 10 Hz, respectively; and
d) DVFC for the same structure as in c). The actuator dynamics (Eq. (3)) is represented by a
pair of high-damped poles, two zeros at the origin and a real pole and the structure is
represented by two zeros at the origin and interlacing low-damped poles and zeros. It is
clearly shown that the resulting root locus has non-collocated system features due to the
influence of the actuator dynamics on the structure dynamics. That is, when the actuator
dynamics are considered, the interlacing pole-zero pattern exhibited by collocated systems is

Fig. 3. Examples of root loci. () poles; () zeros


77
no longer accomplished. On one hand, it is observed that for a structure with a fundamental
frequency of 4 Hz, direct output feedback (DAFC since the actual measurement is the
acceleration) will provide very small relative stability (the distance of the poles to the
imaginary axis in the s-plane) and low damping (Bolton, 1998). However, the inclusion of an
integrator circuit (a pole at the origin for an ideal integrator), which results in DVFC,
improves substantially such properties. On the other hand, for a structure with a higher
fundamental frequency (7 Hz), DAFC provides much better features than DVFC.
Fig. 3 shows the fact that DVFC might not be a good solution and supports the use of CAFC.
It is applied the following phase-lag compensator to the measured acceleration
( )
F
s
C s
s
+
= with
0
. (9)
If = 0 , the control scheme will be DAFC since ( ) 1
F
C s = . If , which means that the
zero of the compensator does not affect the dominant system dynamics, the control scheme
will then be considered DVFC. Parameter has to be chosen according to the closed-loop
poles corresponding to the fundamental frequency of the structure in order to: 1) improve
substantially their relative stability, 2) decrease their angles with respect to the negative real
axis to allow increasing damping, and 3) increase the distance to the origin to allow
increasing natural frequency. Note that increasing values both of the frequency and the
damping result in decreasing the settling time of the corresponding dynamics (Bolton, 1998).
The possible values of that provide the aforementioned features can be bound through
the departure angle at zero gain of the locus corresponding to the fundamental structure
vibration mode. This angle must point to negative values of the real axis. To obtain this
angle, the argument equation of the closed-loop characteristic equation is used. That is, any
point s
1
of a specific trajectory verifies the following equation

( ) ( )
( )
= =
+ + = +

`
1 1
1 1
2 1 with
p
z
n
n
i j
i j
s z s p k k
, (10)
in which n
z
is the number of zeros, n
p
is the number of poles and ( ) +
1 i
s z and
( )
+
1 j
s p
are the angles of vectors drawn from the zeros and poles, respectively, to point s
1
. The
departure angle can be determined by letting s
1
be a point very close to one of the poles of
the fundamental structure vibration mode. As an example, the dominant dynamics are
considered without the direct compensator. Fig. 4 shows the map of zeros and poles under
the aforementioned conditions. Eq. (10) can be written as follows
( ) ( ) ( ) + + + + + + + = + `
1 2 3 4 1 2 3 4 5
2 1 with k k . (11)
If it is considered that the damping of the fundamental vibration mode is
1
0 , the
following assumptions can be done: = =
1 2 3
2 and
5
2 . Therefore, Eq. (11) can
be rewritten as
( ) + + + = `
4 1 2 3 4
2 with k k . (12)
Considering transfer functions (2) and (3), the angles
1
,
2
and
3
are of the following form


78

= = + =

2 2
1 1 1
1 2 3
1 1
atan , atan , and atan
A A
A A A A A A
(13)
Considering transfer functions (2) and (9), the angle
4
is obtained as

=

1
4
atan (14)
Then, by imposing a minimum
4,min
and a maximum
4,max
value of the departure angle
4
of the fundamental structure vibration mode, a couple of values of
4
can be obtained

( ) ( )
= + + + = + + +
4,min 1 2 3 4,min 4,max 1 2 3 4,max
2 + , 2 + (15)
in which it is assumed = 1 k . Therefore, the variation interval of
4
is derived as follows

( ) ( ) ( )
4 4,min 4,max
max 0; , min 2; (16)
and using Eq. (14), the corresponding variation interval of ( )
min max
, is determined. The
final value of must be chosen so that the attractor properties of the zero are focussed on
the fundamental vibration mode.

Fig. 4. Departure angle of the locus corresponding to the fundamental structure vibration
mode (not to scale). () pole; () zeros
3.3 Stability
Stability is the primary concern in any active control system applied to civil engineering
structures, mainly due to safety and serviceability reasons. The control scheme of Fig. 1,
assuming that the nonlinear element f is a saturation nonlinearity, is analysed in this section.
The nonlinear element can be written as
Imaginary
Real
5
2

1 2 3
, ,
2
1
s


79
( ) ( )
( ) ( )
( ) ( ) ( )

=

>

c c c s c
c
s c c s c
K y t y t V K
f y t
Vsign y t y t V K
(17)
where K
c
is the control gain and V
s
is the maximum allowable control voltage to the actuator
(saturation level). The saturation nonlinearity is introduced to avoid actuator force
saturation and to keep the system safe under any excitation and independently of selection
of control parameters. The stability can be studied using the Describing Function (DF) tool
in its basic form (Slotine & Li, 1991). Firstly, the total transfer function of the linear part (Eqs.
(2), (3), (6) and (9)) is obtained (see Fig .1)
( ) ( ) ( ) ( ) ( )
T D A F
G j C j G j G j C j = . (18)
Then, from the root locus of G
T
, the limit gain
,limit c
K for which the closed-loop system
becomes unstable is derived. This is the minimum value of the control gain K
c
(see Eq. (17))
for which at least one of the loci intersects the imaginary axis.
Secondly, the DF, denoted by ( ) , N A , for the nonlinear element is obtained. The DF is the
ratio between the fundamental component of the Fourier series of the nonlinear element
output and a sinusoidal input given by ( ) ( ) = sin x t A t . If the nonlinearity is hard, odd and
single-valued (the case of saturation nonlinearity), the DF depends only on the input
amplitude ( ) ( ) = , N A N A , i.e., it is a real function. The DF for a saturation nonlinearity is
(Slotine & Li, 1991)
( )

=

>

2
2
2
arcsin 1
c
c
K A a
N A
K a a a
A a
A A A
(19)
with =
s c
a V K (see Eq. (17)). The normalised DF ( )
c
N A K (19) is plotted in Fig. 5a as a
function of A a . If the input amplitude is in the linear range ( ) A a , ( ) N A is constant and
equal to the control gain ( ) =
c
N A K . ( ) N A then decreases as the input amplitude increases
when > A a . That is, saturation does not occur for small signals and it reduces the ratio of
the output to input as the input increases.
Thirdly, the extended Nyquist criterion using the DF is applied
( ) ( ) = 1
T
G j N A . (20)
Each solution of Eq. (20) predicts a limit cycle behaviour (self-sustained periodic
undesirable vibration). The total transfer function ( )
T
G j will intersect the real axis at
,limit
1
c
K . With regards to the plot of ( ) 1 N A , it will start at 1
c
K and go to as A
increases. Depending on the value of
c
K , both plots ( )
T
G j and ( ) 1 N A can intersect.
Fig. 5b illustrates this fact. The conclusion is that: if
,limit c c
K K < , the system is
asymptotically stable and goes to zero vibration (no intersection); otherwise, a limit cycle
is predicted (intersection). Such a limit cycle is deduced to be stable by using the limit
cycle stability criterion (Slotine & Li, 1991; Daz & Reynolds, 2010a). The properties of the
limit cycle, frequency and amplitude, can be obtained by Eq. (20) particularised to the
intersection point.


80

Fig. 5. a) DF for the saturation nonlinearity. b) Nyquist diagram of ( )
T
G j and ( ) 1 N A
3.4 Design process
The design process of the control scheme represented in Fig. 1 can be summarised in the
following steps:
Step 1: Identify the actuator G
A
and structure dynamics G.
Step 2: Design the direct compensator (phase-lead) C
D
accounting for the actuator stroke
saturation. This design is carried out following the procedure described in Subsection 3.1.
Step 3: Design the feedback (phase-lag) compensator C
F
to increase the damping and
robustness with respect to stability and performance of the closed-loop system by following
Subsection 3.2.
Step 4: Design the nonlinear element f according to stability and performance. If f is a
saturation nonlinearity, take a saturation value to avoid actuator force overloading and
select a suitable gain K
c
using the root locus method.
4. Implementation on a floor structure
This Section presents the design and practical implementation of an AVC system based on
the procedure presented in Section 3 on an in-service open-plan office floor sited in the
North of England.
4.1 Structure description and modal properties
The test structure is a composite steel-concrete floor in a steel frame office building. A plan
of the floor is shown in Fig. 6a, in which the measurement points used for the experimental
modal analysis (EMA) are indicated. Columns are located along the two sides of the
building (without point numbers) and along the centreline (18-27-end), at every other test
point (TP) location (i.e, 18, 20, 22, etc.). Fig 6b shows a photograph from TP 44 towards TP 28
and Fig. 6c shows a photograph from TP 12 towards TP 01. The EMA of this structure is
explained in detail in (Reynolds et al., 2009). The floor is considered by its occupants to be
quite lively, but not sufficiently lively to attract complaints. Special attention is paid to TP 04
(see Fig. 6a) and its surroundings because it is perceived to be a particularly lively location


81
on the floor. Because the vibration perception is particularly acute at this point, this is the
candidate for the installation of the AVC system.

Fig. 6. Floor structure. a) Plan of the floor (not to scale); () measurement point used in the
EMA. b) View from TP 44 towards TP 28. c) View from TP 12 to TP 01
A multi-input multi-output modal testing is carried out with four excitation points placed at
TPs 04, 07, 31 and 36 and responses measured at all TPs. The artificial excitation is supplied
by APS Dynamics Model 400 electrodynamic shakers (Fig. 7b) and responses are measured
by QA750 force-balance accelerometers (Fig. 7c). Fig 7a shows a photograph of the
multishaker modal testing carried out. Fig. 8 shows the magnitudes of the point accelerance
FRFs acquired. Interestingly, the highest peak occurs at TP 04 at approximately 6.4 Hz,
which is the point on the structure where the response has been subjectively assessed to be
highest. Parameter estimation is carried out using the multiple reference orthogonal
polynomial algorithm already implemented in MEscope suite of software. Fig. 9 shows the
estimated vibration modes which are dominant at TP 04.


82

Fig. 7. a) Multishaker modal testing of the floor structure. b) APS Dynamics Model 400
electrodynamic shaker. c) QA750 force-balance accelerometer

Fig. 8. Magnitudes of the point accelerance FRFs at TP 04, 07, 31 and 36
b)
a) c)
Shaker 1
Shaker 2


83

Fig. 9. Estimated vibration modes prone to be excited by human walking at TP 04. a)
Vibration mode at 6.37 Hz. b) Vibration mode at 9.19 Hz
4.2 System dynamics
The AVC system is placed at TP 04. The floor dynamics at the AVC location (Eq. (2)) and the
actuator dynamics (Eq. (3)) are identified. Using the modal parameters obtained from the
EMA, the transfer function of the floor is modelled considering three vibration modes 3 N =
in the frequency range of 020 Hz

( )
5 2 5 2 4 2
2 2 2
4.49 10 1.74 10 1 10
2.36 1627 2.09 3341 17.59 7738
s s s
G s
s s s s s s

= + +
+ + + + + +
(21)
The inertial actuator is an APS Dynamics Model 400 electrodynamic shaker (operated in
inertial mode and driven using voltage mode) with an inertial mass of 30.4 kg
A
m = (APS
Manual). The actuator model is obtained to be
( )
2
3 2
12600
62.16 728.6 6573
A
s
G s
s s s
=
+ + +
. (22)
The natural frequency of the actuator is estimated as 1.80 Hz and the pole that provides the
low pass property is estimated to be 50.26 rad s = (Eq. (3)).
4.3 Controller design
Compensators C
D
and C
F
are obtained following Section 3. Firstly, C
D
is obtained. From
(22), the transfer function G
d
(Eq. (4)) is derived. By assuming a value of 0.05 m d = , which
is appropriate considering the actual stroke limit of the actuator is 0.075 m, its magnitude
( )
G j (Eq. (5)) is obtained with
14.74 rad s = . The compensator parameters are thus

derived from the optimisation problem (8) in which it is assumed 1.25 rad s
L
= ,
62.83 rad s
U
= and
max max
50.26 = = = . The control parameters are then found to be
6.87 = and 14.34 = .
Secondly, the feedback compensator C
F
(Eq. (9)) is obtained. Taking into account the
dominant dynamics: G
A
, C
D
and the fundamental floor vibration mode of (21) (the first one
of the three considered mode), and restricting the departure angle of the locus
corresponding to the structure vibration mode as ( )
4
180, 225 deg (see Fig. 4), then the
angle corresponding to the zero of the compensator
4
can be bounded. It is obtained


84
( )
4
10.24, 55.24 deg and consequently ( ) 27.9, 223.2 using Eq. (14). A value of 55 =
is chosen since it must be higher than the inferior limit but it should not be so high that the
attractor effect of the zero is focussed on the fundamental vibration mode. The root locus
technique is used here. The root locus of G
T
(Eq. (18)) for CAFC is plotted in Fig. 10. It can be
observed that the linear system might be critically damped for the fundamental floor
vibration mode.

Fig. 10. Root locus of the total transfer function G
T
for CAFC. () pole; () zero; (F) floor; (A)
actuator
Finally, the nonlinear element is chosen as a saturation nonlinearity (17) with 1 V
s
V = ,
which is a convenient value to avoid actuator force overloading at any frequency of the
excitation. Therefore, the stability is guaranteed just by taking a safe control gain
,limit c c
K K < . The limit gain is predicted to be
( )
2
,limit
44.4 V m s
c
K = . Values of damping of
the actuator poles smaller than 0.30 are not recommended (Hanagan, 2005). This damping is
reached for
( )
2
18.6 V m s
c
K = .
A digital computer is used for the on-line calculation of the control signal V. The system
output is sampled with a period 0.001 s t = and the control signal is calculated once every
sampling period. Then, the discrete-time control signal is converted into a zero-order-hold
continuous-time signal. Likewise, the continuous transfer functions of the compensators are
converted to discrete transfer function using the zero-order-hold approximation. Note that
the sampled period used is sufficiently small to not consider the delay introduced by the
digital implementation of the control scheme.
4.4 Results
The AVC system is assessed firstly by carrying out several simulations using different
values of the control gain K
c
. MATLAB/Simulink is used for this purpose. The control
scheme shown in Fig. 1 is simulated using the transfer function models of the compensators
given by Eqs. (6) and (9) with the parameters obtained in Section 4.3 and Eqs. (21) and (22)


85
obtained from FRF identifications. The control scheme is perturbed by a real walking
excitation obtained from an instrumented treadmill (Brownjohn et al., 2004) ( ( ) p t in Fig. 1).
Table 1 shows the results obtained. Two different pacing frequencies (1.58 and 2.12 Hz) are
used in such a way that the first floor vibration mode might be excited by the third or the
fourth harmonic. The results are compared in terms of the maximum transient vibration
value (MTVV) calculated from the 1 s running RMS acceleration and from the vibration dose
value (VDV) obtained from the total period of the excitation (BS 6472, 2008). The BS 6841 W
b

weighted acceleration is used for both measures (BS 6841, 1987). The results predicted that
the AVC are quite insensitive to the gain value. The reduction in vibration is approximately
60 % for slow walking (1.58 Hz) and 53 % for fast walking (2.12 Hz) in terms of MTVV. The
results in terms of the VDV provide similar reductions as for the MTVV.

Control gain (V/(m/s
2
)) Uncontrolled 5 10 15 20
Walking at 1.58 Hz
MTVV
(1)
(m/s
2
) 0.030 0.0122 0.0115 0.0115 0.0116
Reduction MTVV (%) 59 62 62 61
VDV
(2)
(m/s
1.75
) 0.063 0.029 0.027 0.027 0.027
Reduction VDV (%) 55 58 58 57
Walking at 2.12 Hz
MTVV (m/s
2
) 0.033 0.0154 0.0150 0.0156 0.0160
Reduction MTVV (%) 53 55 53 52
VDV (m/s
1.75
) 0.073 0.034 0.032 0.033 0.033
Reduction VDV (%) 54 56 55 55
Table 1. Simulation performance assessment for the floor structure using several control
gains and walking excitation.
(1)
Maximum Transient Vibration value defined as the
maximum value of 1s running RMS acceleration.
(2)
Cumulative Vibration Dose Value
Actual walking tests are carried out on the test structure using the same walking excitation
frequencies as in the simulations. The walking path consists of walking from TP 01 to TP 09
and then back from TP 09 to TP 01 (see Fig. 6). A gain of
( )
2
15 V m s
c
K = is found to give
good performance so that is used in the experiments. Fig. 11 shows BS 6841 W
b
weighted
response time histories (including the 1s RMS and the cumulative VDV), uncontrolled and
controlled, for a pacing frequency of 1.58 Hz, which is controlled using a metronome set to
95 beats per minute (bpm). The MTVV is reduced from
2
0.031 m s to
2
0.010 m s , a
reduction of 68 %, and the VDV is reduced from
1.75
0.050 m s to
1.75
0.019 m s , a reduction
of 62 %. The same test is carried out for a pacing frequency of 2.12 Hz (127 bpm). The
achieved reduction in terms of the MTVV is now 52 %, from
2
0.033 m s to
2
0.016 m s ,
and in terms of the VDV is 51 %, from
1.75
0.057 m s to
1.75
0.028 m s . The experimental
reductions agreed very well with the numerical predictions (see Table 1).
Continuous whole-day monitoring has been carried out to assess the vibration reduction
achieved by the AVC system. The acceleration is measured from 6:00 am to 6:00 pm during
four working days, two without and two with the AVC system. The R-factor is used to
quantify the vibration reduction. This factor is defined as the ratio between the 1 s running
RMS of the BS 6841 W
b
weighted acceleration response and 0.005 m/s
2
(Wyatt, 1989). Fig. 12
shows the percentage of time during which the R-factor is over 1, 2, 3, 4 and 5. The values


86

Fig. 11. Experimental results on the floor structure. Walking at 1.58 Hz (95 bpm). a)
Uncontrolled
2
MTVV 0.031 m s = and
1.75
VDV 0.050 m s = . b) Controlled
2
MTVV 0.010 m s = and
1.75
VDV 0.019 m s =

Fig. 12. Whole-day monitoring: percentage of time of exceedance of R-factors


87
shown in the Fig. are the mean values between the two corresponding days. It is observed
that the time for which the R-factor is over 1 is reduced by 60 % and the time for which it is
over 4 is significantly reduced by over 97 %. Note that the second reduction is very
important since an R-factor of 4 is a commonly used vibration limit for a high quality office
floor (Pavic & Willford, 2005). Hence, these results clearly illustrate the effectiveness of the
AVC system designed. In addition, the cumulative VDV is also calculated for the same
exposure to vibration and using the W
b
weighted acceleration. The VDV obtained when the
AVC system is disconnected is
1.75
0.162 m s whereas such a value is
1.75
0.101 m s when
the system is engaged. The reduction achieved is almost 40 %. Note that the VDV is much
more strongly influenced by vibration magnitude than duration (BS 6472, 2008). This fact
results in less vibration reduction in terms of the VDV than using the MTVV.
5. Implementation on a footbridge
This Section presents the design and practical implementation of an AVC system based on
the procedure presented in Section 3 on an in-service footbridge sited in Valladolid (Spain).
5.1 Structure description and modal properties
The test structure, sited in Valladolid (Spain), is a footbridge that creates a pedestrian link
over The Pisuerga River between the Science Museum and the city centre (see Fig. 13). This
bridge, built in 2004, is a 234 m truss structure composed of four spans: three made of
tubular steel beams and one made of white concrete, all of them with a timber walkway. The
main span (Span 3 in Fig. 1), post-tensioning by two external cable systems (transversal and
longitudinal), is 111 m, the second span (Span 2 from this point onwards) is 51 m and the
other two spans are shorter and stiffer (Gmez, 2004). The external cable systems of Span 3
has both aesthetical reasons (the original design by the architect Jos Rafael Moneo was
based on the form of a fish basket) and structural reasons (making Span 3 stiffer).
Because of its slenderness, this footbridge, especially Span 2, represents a typical lightweight
structure sensitive to dynamic excitations produced by pedestrians. Annoying levels of
vibration are sometime perceived in Span 2. Special attention is paid to the point of maximum
amplitude of the first bending mode (close to mid-span) since the vibration perception is acute
at this point, particularly when runners cross the bridge. Therefore, it is decided to study the
dynamic properties of this span and implement the AVC system at that point.

Fig. 13. General view of the Valladolid Science Museum Footbridge
Science
Museum
Span 2
Span 3,
Main span Post-tensioning tendons
The Pisuerga River


88
The operational modal analysis of Span 2 is carried out in order to obtain the natural
frequencies, damping ratios and modal shapes of the lower vibration modes. The analysis is
carried out with five roving and two reference accelerometers. Preliminary spectral analyses
and time history recordings indicates that the vertical vibration is considerably higher than
the horizontal one, thus, only vertical response measurements are performed. A
measurement grid of 3 longitudinal lines with 9 equidistant test points is considered,
resulting in 27 test points. Five setups with an acquisition time of 720 seconds and a
sampling frequency of 100 Hz are recorded. Thus, it is expected to successfully identify
vibration modes up to 30 Hz. The modal parameter estimation is carried out using the
ARTeMIS suite of software. In particular, frequency domain methods (Frequency domain
decomposition-FDD, enhanced frequency domain decomposition-EFDD and curve-fit
frequency domain decomposition-CFDD) are used. Table 2 shows the modal parameters
estimated through the modal analysis for the first four vibration modes. Fig. 14 shows the
corresponding estimated modal shapes.

Mode 1 Mode 2 Mode 3 Mode 4
FDD
Frequency (Hz) 3.516 6.250 7.373 9.351
Damping ratio (%)
EFDD
Frequency (Hz) 3.506 6.278 7.386 9.365
Damping ratio (%) 0.7221 0.4167 0.6571 0.5528
CFDD
Frequency (Hz) 3.508 6.274 7.389 9.367
Damping ratio (%) 0.7984 0.2599 0.4319 0.3869
Table 2. Natural frequencies and damping ratios identified by the operational modal
analysis of the footbridge

1
st
bending mode: 3.5 Hz

1
st
torsional mode: 6.3 Hz

2
nd

3
rd
Fig. 14. Estimated modal shapes for the footbridge structure


89
5.2 System dynamics
The FRF between the structure acceleration and the input force is obtained at the middle of
the transversal steel beam sited closest to the point of maximum value of the first vibration
mode (close to mid-span). A chirp signal with frequency content between 1 and 15 Hz is
used to excite the first vibration modes. The force is generated by an APS Dynamics Model
400 electrodynamic (as was used for the floor) and the structure acceleration is measured by
a piezoelectric accelerometer (as those used for the modal analysis). The force induced by
the shaker is estimated by measuring the acceleration of the inertial mass and multiplying
this by the magnitude of the inertial mass (30.4 kg). Thus, the structure dynamics for the
collocated case between the acceleration (output) and the force (input) can be identified
from model (2) as follows
( )
5 2 5 2
2 2
5.40 10 5.85 10
0.3079 483.6 0.5887 3451
s s
G s
s s s s

= +
+ + + +
(23)
The vibration modes at 6.3 and 7.4 Hz (see Fig. 14) are not clearly observed and they are not
included into this model. The same shaker that is used to obtain the FRF for the structure is
used as inertial actuator. The transfer function between the output force and the input
voltage is the one given by Eq. (22).
5.3 Controller design
The design process shown in Section 3 is now followed. The direct compensator of the form
of (6) is assumed. Again, a maximum stroke for harmonic excitation of d = 0.05 m is
considered in the design. The compensator parameters are obtained from the optimisation
problem (8) using the parameter used in Subsection 4.3. The controller parameters are found
to be 5.6 = and 24.6 = . These parameters are selected in such a way that the likelihood
of stroke saturation is reduced significantly. The stroke saturation leads to collisions of the
inertial mass with its stroke limits, imparting highly undesirable shocks to the structure and
possibly causing damage to the actuator.
Once the direct compensator is designed, the feedback one (9) is designed considering the
dynamics of the actuator, structure and the direct compensator dynamics. As in Subsection
4.3, the departure angle of the locus corresponding to the structure vibration mode is
restricted to ( )
4
180, 225 deg (see Fig. 4). It is obtained 35.5 . A value of 50 = is
finally chosen. The root locus technique is now used. The root locus of the total transfer
function of the linear part (18) is plotted in Fig. 15 (only the dominant dynamics are shown).
It is observed that a couple of branches in the root locus corresponding to the actuator
dynamics go to the right-half plane provoking unstable behaviour in the actuator. The gain
for which the control system is unstable is the limit gain
( )
2
,limit
82 V m s
c
K = . Finally, a
gain of
( )
2
,limit
40 V m s
c
K = is chosen. The saturation level is set to 1 V
s
V = .
5.4 Results
Once both compensators and the control parameters are selected, simulations are carried out
in order to assess the AVC performance. MATLAB/Simulink is again used for this purpose.
Table 3 shows controlled acceleration response for walking and running excitation.
Moreover, the AMD displacement estimation is included.


90

Fig. 15. Root locus of the total transfer function G
T
for CAFC on the footbridge. () pole; ()
zero; (F) footbridge; (A) actuator

A
F
F


91

Uncontrolled
(m/s
2
)
Controlled
(m/s
2
)
Reduction
(%)
Mass displacement
(m)
Walking at 1.75 Hz 0.39 0.04 89 0.034
Running at 3.50 Hz 6.16 3.75 40 0.022

Table 3. Simulation performance assessment for the footbridge using the peak acceleration
for walking and running excitation
Walking and running tests are carried out to assess the efficacy of the AVC system. The
walking tests consist of walking at 1.75 Hz such that the first vibration mode of the structure
(3.5 Hz) could be excited by the second harmonic of walking. A frequency of 3.5 Hz is used
for the running tests so that the structure is excited by the first harmonic of running. The
walking/running tests consisted of walking/running from one end of Span 2 to the other
and back again. The pacing frequency is controlled using a metronome set to 105 beats per
minute (bpm) for 1.75 Hz and to 210 bpm for 3.5 Hz. Each test is repeated three times.

Uncontrolled Controlled Reduction (%)
Walking at 1.75 Hz
Peak acceleration (m/s
2
) 0.41 0.16 70
MTVV
(1)
(m/s
2
) 0.21 0.06 67
Running at 3.50 Hz
Peak acceleration (m/s
2
) 3.34 1.19 64
MTVV (m/s
2
) 2.20 0.69 68

Table 4. Experimental performance assessment for walking and running excitation.
(1)

Maximum Transient Vibration value defined as the maximum value of 1s running RMS
acceleration
The results are compared by means of the maximum peak acceleration and the MTVV
computed from the 1 s running RMS acceleration. Table 4 shows the result obtained for the
uncontrolled and controlled case. It is observed that the AMD designed (with a moving
mass of 30 kg) performs well for both excitations, achieving reductions of approximately 70
%. Fig. 16 shows the response time histories (including the 1 s RMS) uncontrolled and
controlled for a walking test. Fig. 17 shows the same plots for a running test.


92

Fig. 16. Walking test on the footbridge. a) Uncontrolled
2
MTVV 0.207 m s = . b) Controlled
2
MTVV 0.067 m s =
a)
b)


93

Fig. 17. Running test on the footbridge. a) Uncontrolled
2
MTVV 2.198 m s = . b) Controlled
2
MTVV 0.773 m s =
a)
b)


94
6. Conclusion
The active cancellation of human-induced vibrations has been considered in this chapter.
Even velocity feedback has been used previously for controlling human-induced vibrations,
it has been shown that this is not a desirable solution when the actuator dynamics influence
the structure dynamics. Instead of using velocity feedback, here, it is used a control scheme
base on the feedback of the acceleration (which is the actual measured output) and the use
of a first-order compensator (phase-lag network) conveniently designed in order to achieve
significant relative stability and damping. Note that the compensator could be equivalent to
an integrator circuit leading to velocity feedback, depending on the interaction between
actuator and structure dynamics. Moreover, the control scheme is completed by a phase-
lead network to avoid stroke saturation due to low-frequency components of excitations and
a nonlinear element to account for actuator overloading. An AVC system based on this
control scheme and using a commercial inertial actuator has been tested on two in-service
structures, an office floor and a footbridge.
The floor structure has a vibration mode at 6.4 Hz which is the most likely to be excited. This
mode has a damping ratio of 3% and a modal mass of approximately 20 tonnes. Reductions
of approximately 60 % have been observed in MTVV and cumulative VDV for controlled
walking tests. For in-service whole-day monitoring, the amount of time that an R-factor of 4
is exceeded, which is a commonly used vibration limit for high quality office floor, is
reduced by over 97 %. The footbridge has a vibration mode at 3.5 Hz which is the most
likely to be excited. This mode has a damping ratio of 0.7 % and a modal mass of
approximately 18 tonnes. Reductions close to 70 % in term of the MTVV has been achieved
for walking and running tests.
It has been shown that AVC could be a realistic and reasonable solution for flexible
lightweight civil engineering structures such as light-weight floor structure or lively
footbridges. In these cases, in which low control forces are required (as compared with other
civil engineering applications such as high-rise buildings or long-span bridges), electrical
actuators can be employed. These actuators present advantages with respect to hydraulic
ones such as lower cost, maintenance and level of noise. However, AVC systems for human-
induced vibrations needs much further research and development to jump into building and
construction technologies considered by designers. With respect to passive systems, such as
TMDs, cost is still the mayor disadvantage. However, it is expected that this technology will
become less expensive and more reasonable in the near future. Research projects involving
the development of new affordable and compact actuators for human-induced vibration
control are currently on the go (Research Grant EP/H009825/1, 2010).
7. Acknowledgment
The author would like to acknowledge the financial support of Universidad de
Castilla-La Mancha (PL20112170) and Junta de Comunidades de Castilla-La Mancha
(PPII11-0189-9979. The author would like to thank his colleagues Dr. Paul Reynolds and Dr
Donald Nyawako from the University of Sheffield, and Mr Carlos Casado and Mr Jess de
Sebastin from CARTIF Centro Tecnolgico for their collaboration in works presented in
this chapter.


95
8. References
APS. Instruction Manual Electro-Seis Model 400 Shaker, APS Dynamics, USA, available from
http://www.apsdynamics.com
Bachmann, H. (1992). Case studies of structures with man-induced vibrations. Journal of
Structural Engineering, Vol.118, No.3, pp. 631-647, ISNN 0733-0445
Bachmann, H. (2002). Lively footbridgesa Real Challenge, Proceedings of the International
Conference on the Design and Dynamic Behaviour of Footbridges, OTUA, Paris, France,
November 20-22
Balas, M.J. (1979). Direct velocity feedback control of large space structures, Journal of
Guidance and Control, Vol.2, No.3, pp. 25253
Bolton, W. (1998). Control engineering, Logman, ISBN 978-0-582-32773-3, United Kingdom
Brownjohn, J.M.W., Pavic, A. & Omenzetter, P. (2004). A spectral density approach for
modelling continuous vertical forces on pedestrian structures due to walking,
Canadian Journal of Civil Engineering, Vol.31, No.1, pp. 6577, ISSN 0315-1468
BS 6841. (1987). Measurement and evaluation of human exposure to whole-body mechanical
vibration and repeated shock, British Standards Institution, ISBN 0-580-16049-1,
United Kingdom
BS 6472. (2008). Guide to evaluation of human exposure to vibration in buildings. Part 1: Vibration
sources other than blasting, British Standards Institution, ISBN 978-0-580-53027-2,
United Kingdom
Caetano, E., Cunha, A., Moutinho, C. & Magalhes, F. (2010) Studies for controlling human-
induced vibration of the Pedro e Ins footbridge, Portugal. Part 2: Implementation
of tuned mass dampers, Engineering Structures, Vol.32, pp. 10821091, ISSN 0141-
0296
Chung, L.Y. & Jin, T.G. (1998). Acceleration feedback control of seismic structures,
Engineering Structures, Vol.20, No.1, pp. 6274, ISSN 0141-0296
Daz, I.M. & Reynolds, P. (2010a). On-off nonlinear active control of floor vibrations,
Mechanical Systems and Signal Processing, 24: 17111726, ISSN 0888-3270
Daz, I.M. & Reynolds, P. (2010b). Acceleration feedback control of human-induced floor
vibrations, Engineering Structures, Vol.32, No.1, pp. 163173, ISSN 0141-0296
Ebrahimpour, A. & Sack, R.L. (2005). A review of vibration serviceability criteria for floor
structures, Computers and Structures, Vol.83, pp. 248894, ISSN 0045-7949
FIB-Bulletin 32. (2005). Guidelines for the design of footbridges, International Federation for
Structural Concrete, Lausanne, Switzerland
Gmez, M. (2004). A new and unusual cable-stayed footbridge at Valladolid (Spain).
Steelbridge 2004: Symposium International sur les Ponts Metlliques, Milau, France,
June, pp. 23-25
Hanagan, L.M. & Murray, T.M. (1997) Active control for reducing floor vibrations, Journal of
Structural Engineering, Vol.123, No.11, pp. 14971505, ISSN 0733-9445
Hanagan, L.M., Raebel, C.H. & Trethway, M.W. (2003a). Dynamic measurements of in-place
steel floors to assess vibration performance, Journal of Performance of Constructed
Facilities, Vol.17, pp. 126135, ISSN - 0887-3828


96
Hanagan, L.M., Murray, T.M. & Premaratne, K. (2003b). Controlling floor vibration with
active and passive devices, The Shock and Vibration Digest, Vol.35, No.5, pp. 34765,
ISSN 0583-1024
Hanagan, L.M. (2005). Active floor vibration system, United States Patent 6874748
Moutinho, C., Cunha, A. & Caetano, E. (2010). Analysis and control of vibrations in a stress-
ribbon footbridge, Structural Control and Health Monitoring, doi: 10.1002/stc.390
Nyawako, D. & Reynolds, P. (2007) Technologies for mitigation of human-induced vibration
in civil engineering structures, The Shock and Vibration Digest, Vol.36, No.(6), pp.
46593, ISSN 0583-1024
Occhiuzzi, A., Spizzuoco, M. & Ricciardelli, F. (2008). Loading models and response control
of footbridges excited by running pedestrians, Structural Control and Health
Monitoring, Vol.15, pp. 349368, ISSN 1545-2263
Pavic, A. & Willford, M. (2005). Appendix G in Post-tensioned concrete floors design handbook
Technical Report 43, Concrete Society, Slough, United Kingdom
Preumont, A. (1997). Vibration Control of Active Structures: An introduction, Kluwer Academic,
Dordrecht, ISBN 1-4020-0496-9, The Netherlands
Reiterer, M. & Ziegler, F. (2006). Control of pedestrian-induced vibrations of long-span
bridges, Structural Control and Health Monitoring, Vol.13, pp. 10031027, ISSN 1545-
2263
Research Grant EP/H009825/1. (2010). Active control of human-induced vibration, PI: Dr Paul
Reynolds, Engineering and Physical Sciences Research Council, 20102012, United
Kingdom
Reynolds, P., Daz, I.M. & Nyawako, D.S. (2009). Vibration testing and active control of an
office floor, Proceedings of the 27th International Modal Analysis Conference, Orlando,
Florida, USA
Setareh, M. & Hanson, R.D. (1992). Tuned mass damper to control floor vibration from
humans, Journal of Structural Engineering, Vol.118, No.3, pp. 74162, ISSN 0733-9445
Setareh, M. (2002). Floor vibration control using semi-active tuned mass dampers, Canadian
Journal of Civil Engineering, Vol.29, No.1, pp. 7684, ISSN 0315-1468
Slotine, J.J. & Li, W. (1991). Applied non linear control, Prentice-Hall, Chapter 5, ISBN 013-
040890-5, USA
Wyatt, T.A. (1989). Design guide on the vibration of floors, The Steel Construction Institute,
ISBN 1-870004-34-5, United Kingdom
5
Control Strategies for Vehicle Suspension
System Featuring Magnetorheological
(MR) Damper
Min-Sang Seong
1
, Seung-Bok Choi
1
and Kum-Gil Sung
2

1
Inha University
2
Yeungnam College of Science and Technology
Korea
1. Introduction
Vehicle suspension is used to attenuate unwanted vibrations from various road conditions.
So far, three types of suspension system have been proposed and successfully implemented;
passive, active and semiactive. Though the passive suspension system featuring oil damper
provides design simplicity and cost-effectiveness, performance limitations are inevitable
due to the lack of damping force controllability. On the other hand, the active suspension
system can provides high control performance in wide frequency range. However, this type
may require high power sources, many sensors and complex actuators such as servovalves.
Consequently, one way to resolve these requirements of the active suspension system is to
adopt the semiactive suspension system. The semiactive suspension system offers a
desirable performance generally enhanced in the active mode without requiring large power
sources and expensive hardware.
One of very attractive and effective semiactive vehicle suspension systems is to utilize
magnetorheological (MR) fluid. MR fluids are currently being studied and implemented as
actuating fluids for valve systems, shock absorbers, engine mounts, haptic systems,
structure damper, and other control systems. The rheological properties of MR fluids are
reversibly and instantaneously changed by applying a magnetic field to the fluid domain.
Recently, a very attractive and effective semi-active suspension system featuring MR fluids
has been researched widely. Carlson et al., 1996 proposed a commercially available MR
damper which is applicable to on-and-off-highway vehicle suspension system. They
experimentally demonstrated that sufficient levels of damping force and also superior
control capability of the damping force by applying control magnetic field. Spencer Jr. et al.,
1997 proposed dynamic model for the prediction of damping force of a MR damper. They
compared the measured damping forces with the predicted ones in time domain. Kamath et
al., 1998 proposed a semi-active MR lag mode damper. They proposed dynamic model and
verified its validity by comparing the predicted damping force with the measured one. Yu et
al., 2006 evaluated the effective performance of the MR suspension system by road testing.
Guo & Hu, 2005 proposed nonlinear stiffness model of a MR damper. They proposed
nonlinear stiffness model and verified it using simulation and experiment. Du et al., 2005

Vibration Analysis and Control New Trends and Developments 98
proposed H-infinity control algorithm for vehicle MR damper and verified its effectiveness
using simulation. Shen et al., 2007 proposed load-levelling suspension with a
magnetorheological damper. Pranoto et al., 2005 proposed 2DOF-type rotary MR damper
and verified its efficiencies. Ok et al., 2007 proposed cable-stayed bridges using MR
dampers and verified its effectiveness using semi-active fuzzy control algorithm. Choi et al.,
2001 manufactured an MR damper for a passenger vehicle and presented a hysteresis model
for predicting the field-dependent damping force. Hong et al., 2008 derived a
nondimensional Bingham model for MR damper and verified its effectiveness through
experimental investigation. Yu et al., 2009 developed human simulated intelligent control
algorithm and successfully applied it to vibration control of vehicle suspension featuring
MR dampers. Seong et al., 2009 proposed hysteretic compensator of MR damper. They
developed nonlinear Preisach hysteresis model and hysteretic compensator and
demonstrated its damping force control performance.
As is evident from the previous research work, MR damper is very effective solution for
vibration control of vehicle suspension system. So in this chapter, we formulate various
vibration control strategies for vibration control of MR suspension system and evaluate their
control performances. In order to achieve this goal, material characteristics of MR fluid are
explained. Then the MR damper for vehicle suspension system is designed, modelled and
manufactured. The characteristics of manufactured MR damper are experimentally
evaluated. For vibration control, the quarter vehicle suspension system featuring MR
damper is modelled and constructed. Then, various vibration control strategies such as
skyhook control, PID control, LQG control, H control, Sliding mode control, moving
sliding mode control and fuzzy moving sliding mode control are formulated. Finally,
control performances of the proposed control algorithms are experimentally evaluated and
compared.
2. Suspension modelling
2.1 MR fluid
Since Jacob Rabinow discovered MR fluid in the late 1940s, of which yield stress and
viscosity varies in the presence of magnetic field, various applications using MR fluid have
been developed such as shock absorbers, clutches, engine mounts, haptic devices and
structure dampers, etc (Kim et al., 2002). Physical property changes of MR fluid are resulted
from the chain-like structures between paramagnetic MR particles in the low permeability
solvent. At the normal condition, MR fluid shows the isotropic Newtonian behavior because
the MR particles move freely as shown in Fig. 1 (a). However, when the magnetic field
applied to the MR fluid, MR fluid shows the anisotropic Bingham behavior and resist to
flow or external shear force because the MR particles make a chain structure as shown in
Fig. 1 (b). From this property, force or torque of application devices can be easily controlled
by the intensity of the magnetic field.
2.2 MR damper
The schematic configuration of the cylindrical type MR damper proposed in this work is
shown in Fig. 2. The MR damper is composed of the piston, cylinder and gas chamber. The
floating piston between the cylinder and the gas chamber is also used in order to
compensate for the volume induced by the motion of the piston. Also the gas chamber

Control Strategies for Vehicle Suspension System Featuring Magnetorheological (MR) Damper 99

(a) no magnetic field applied

(b) magnetic field applied
Fig. 1. Phenomenological behavior of MR fluid
which is filled with nitrogen gas acts as an accumulator for absorbing sudden pressure
variation of lower chamber of the MR damper induced by the rapid motion of the piston. The
MR damper is divided into the upper and lower chambers by piston, and it is fully filled with
the MR fluid. By the motion of the piston, the MR fluid flows through the annular duct
between inner and outer piston from one chamber to the other. The magnetic poles in the
piston head is placed to control the yield stress of the MR fluid by supplying current to the
coil. In order effectively to generate the magnetic field in the magnetic pole, the outer cylinder
and both ends of inner piston are made of ferromagnetic substance, while the center of the
inner piston is a paramagnetic substance. In the absence of a magnetic field, the MR damper
produces a damping force caused only by fluid viscous resistance. However, if a certain level
of magnetic field is supplied to the MR damper, the MR damper produces an additional
damping force owing to the yield stress of the MR fluid. This damping force of the MR
damper can be continuously tuned by controlling the intensity of the magnetic field.
In order to simplify the analysis of the MR damper, it is assumed that the MR fluid is
incompressible and that pressure in one chamber is uniformly distributed. The pressure drops
due to the geometric shape of the annular duct and the fluid inertia are assumed to be negligible.
For laminar flow in the annular duct, the fluid resistance is given by (Liu et al., 2006; White, 1994)
R
c
= 8
L
n_R
c
4
-R
i
4
-
[R
c
2
-R
i
2
2
ln(R
c
R
i
)
_
(1)
Particle
Base Oil
Magnetic Pole
N
S


(a) MR damper

(b) piston (3-D view)
Fig. 2. Schematic configuration of the proposed MR damper
where is the viscosity of the MR fluid and I is the length of the annular duct. R
o
and R

are the inner radius of the outer piston and outer radius of the inner piston respectively.
By assuming that the gas does not exchange much heat with its surroundings, and hence
considering its relation as adiabatic variation, the compliance of the gas chamber is
obtained by
C
g
=
v
0
P
0
k
(2)
where I
0
and P
0
are the initial volume and pressure of the gas chamber respectively, and
is the specific heat ratio. On the other hand, the pressure drop due to the increment of the
yield stress of the MR fluid is given by
P
MR
= 2
cL
p
h

(B)

(3)
where c is a coefficient that depends on flow velocity profile and has a value range from 2.0
to 3.0, I
p
is the length of the magnetic pole, is the gap of the annular duct, and
(B) is the

yield stress caused by the magnetic flux density B. Therefore, the damping force of the
proposed MR damper can be written as

F
= k
c
x
p
+c
c
x
p
+ F
MR

(4)
where

Parameter Value
Duct Length (L) 82mm
Piston Area (A
p
) 1661.90mm
2
Piston Rod Area (A
r
) 380.13mm
2

Duct Width (b) 123.53mm
Maximum Stroke 164mm
Table 1. Design parameters of the MR damper
k
c
=
A
r
2
C
g
, c
c
= (A
p
A
)
2
R
c

F
MR
= (A
p
A
)P
MR
sgn(x
p
)

(5)
where x
p
and x
p
are the piston displacement and velocity respectively, A
p
and A
represent
the piston and piston rod areas respectively, and sgn() is a signum function.

Fig. 3. Photograph of the manufactured MR damper
The photograph of the manufactured MR damper with optimally obtained design
parameters are shown in Fig. 3. The principal design parameters of the manufactured MR
damper, which can be applied to a mid-sized commercial passenger vehicle, are presented

in Table 1. Fig. 4 presents the measured and analysed damping force F
D
characteristics of the
MR damper with respect to the piston velocity at various magnetic fields. This is obtained
by calculating the maximum damping force at each velocity. The piston velocity is changed
by increasing the excitation frequency from 0.5 to 4.0Hz, while the excitation amplitude is
kept constant _20mm. This type of plot is frequently used in the damper manufacturing
industry to evaluate the level of damping performance. It is clearly observed that the
damping force is increased as the magnetic field increases, as expected. The damping force
response of the MR damper is measured as shown in Fig. 5. It can be found that the time
constant is about 23ms, which is obtained by inspecting the required time when the
damping force reaches 63.2percent of its final steady state value.

-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6
-2000
-1500
-1000
-500
0
500
1000
1500
2000
0.0A (measured)
1.0A (measured)
2.0A (measured)
0.0A (analyzed)
1.0A (analyzed)
2.0A (analyzed)

D
a
m
p
i
n
g

F
o
r
c
e

(
N
)
Velocity (m/s)

Fig. 4. Field-dependent damping force of the MR damper

0.00 0.05 0.10 0.15 0.20 0.25
400
800
1200
1600
2000
0
1
2
3
Damping Force
Current
D
a
m
p
i
n
g

F
o
r
c
e

(
N
)
Time (sec)

A
p
p
l
i
e
d

C
u
r
r
e
n
t

(
A
)

Fig. 5. Damping force responses of the MR damper

2.3 Quarter vehicle MR suspension
In order to investigate the effectiveness of vibration control algorithm, a quarter vehicle MR
suspension system is constructed as shown in Fig. 6. It shows that the proposed quarter
vehicle suspension model with the MR damper has two degrees of freedom. The spring for
the suspension is assumed to be linear and the tire is also modelled as linear spring
component. From the mechanical model, dynamic equation of system considering time
constant can be expressed as follows:

Fig. 6. Mechanical model of a quarter vehicle MR suspension system
m
s
z
s
= k
s
z
s
c
s
z
s
+k
s
z
u
+ c
s
z
u
F
MR

m
u
z
u
= k
s
z
s
+c
s
z
s
(k
t
+ k
s
)z
u
c
s
z
u
+ F
MR
+ k
t
z
(6)

F
MR
=
1
:
MRD
F
MR
+
1
:
MRD
u

Parameter Value
Sprung Mass (m
s
) 373.5kg
Unsprung Mass (m
u
) 40kg
Stiffness Coefficient (k
s
) 27358N/m
Tire Stiffness Coefficient (k
t
) 211625N/m
Damping Coefficient (c
s
) 570Ns/m
Table 2. System parameters of the quarter vehicle suspension system
where
MR
is time constant of the MR damper, k
s
is the total stiffness coefficient of the
suspension, including the effective stiffness k
c
of the MR damper in equation (4), c
s
is the
damping coefficient of the suspension and is assumed to be equal to c
c
, k
t
is the vertical
stiffness of the tire, and z
s
, z
u
and z
are the vertical displacements of sprung mass,

unsprung mass and road excitation respectively. The state space equation of proposed
quarter vehicle suspension can be expressed using dynamic equation (Lee et al., 2011):
x = Ax + Bu +Lz
, y = Cx

(7)
u
m
s
m
s
z
u
z
r
z
s
k
s
c
d
F
t
k

where
x = |z
s
z
s
z
u
z
u
F
MR
]
1

A =
l
l
l
l
l
l
l
u 1 u u u
k
s
m
s
,
c
s
m
s
,
k
s
m
s
,
c
s
m
s
,
1
m
s
,
u u u 1 u
k
s
m
u
,
c
s
m
u
,
(k
t
+ k
s
)
m
u
,
c
s
m
u
,
1
m
u
,
u u u u
1

MR
,
1
1
1
1
1
1
1
, (8)
B = |u u u u
1

MR
, ]
1
, L = j
u u u
k
t
m
u
, u
[
1
,
C = |
1 u u u u
]
The system parameters of the quarter vehicle MR suspension system are chosen on the
basis of the conventional suspension system for a mid-sized passenger vehicle, and listed
in Table 2.
3. Control strategies
In order to evaluate vibration control performance of the quarter vehicle MR suspension
system, various control strategies are formulated and experimentally implemented.
3.1 Skyhook controller
Skyhook controller is simple but very effective control algorithm. It is well known that the
logic of the skyhook controller is easy to implement in the real field. The principle of
skyhook control is to design the active or semiactive suspension control so that the sprung
mass is linked to the sky in order to reduce the vertical oscillations of the sprung mass. Fig. 7
shows the conceptual scheme of skyhook controller for vehicle suspension system. The
desired damping force is set by

u = C
sk
z
s

(9)

Fig. 7. Scheme of the skyhook controller for vehicle suspension system

where C
sk
is the control gain, which physically indicates the damping coefficient. In this
work, the value is chosen as 2470 using trial-and-error method.
3.2 LQG controller
Optimal control of a linear system with respect to the quadratic objective function under
incomplete measurements corrupted by white Gaussian noise is generally referred to as the
linear quadratic Gaussian (LQG) problem. The optimal control is a linear function of the
state estimates obtained from the Kalman-Bucy filter. The LQR (linear quadratic regulation)
is a state feedback problem, whereas LQG is an output feedback problem, which is more
realistic (Bahram & Michael, 1993). Fig. 8 shows the block diagram of an LQG controller. In
this study, the control input is formulated as follows:
u = K
Lu
x

(10)
where x is the estimated state. Control gain K
Lu
is set as
|
1248.S 11Su.S 4121.8 1S.S u.4
] in this work.
3.3 H controller
In reality, the sprung mass of the vehicle is varied by the loading conditions such as the
number of riding persons and payload. And it makes the pitch and roll mass moment of
inertia to be changed. Therefore, in order to successfully control the vibration, a robust control
algorithm is required by considering the parameter variations of the system. From the
structured suspension model, a desirable damping force required to effectively suppress the
vibration is determined by adopting H controller. Fig. 9 shows the nominal plant, perturbed
plant, shaped plant and loop gain graph. As shown in Fig., loop gain and perturbed plant are
well matched and it means loop perturbation is successfully constructed.
For reducing the steady state error and suppressing the effectiveness of disturbance,
weighting function can be designed as follows (Choi et al., 1999; 2002):
W = 18u
s+88
s

(11)
For the design of controller, design index is set as 2.99. Fig. 10 shows the sensitivity and
complementary sensitivity of the closed loop system. Therefore, control gain can be
calculated as follows:
u = K
H
x, K
H
=
N(s)
(s)

(12)
where N(s) = 1.4Sc2s
6
2.29c4s
5
1.82c6s
4
1.12c8s
3
2.1uc9s
2
+ S.2uc1us
6.91c6, (s) = s
7
+ 92.uSs
6
+ 8.S4cSs
5
+S.8ScSs
4
+ S.47c6s
3
+S.48c7s
2
+7.S1c2s.
3.4 Sliding mode controller (SMC)
Fig. 11 shows the conceptual scheme of sliding mode control algorithm. After the initial
reaching phase, the system states slides along the sliding surface. The first step to formulate
the SMC is to design a stable sliding surface. The stable sliding surface for the control
system is defined as follows:
s = c
1
x
1
+ c
2
x
2
+c
3
x
3
+c
4
x
4
+x
5

(13)


Fig. 8. Block diagram of LQG controller

0.1 1 10 100 1000
-200
-150
-100
-50
0
50

S
i
n
g
u
l
a
r

V
a
l
u
e

(
d
B
)
Frequency (rad/s)
Norminal Plant
Perturbed Plant
Shaped Plant
Loop Gain

Fig. 9. Singular value plots of the quarter vehicle MR suspension system

0.1 1 10 100 1000
-100
-50
0
50

S
i
n
g
u
l
a
r

V
a
l
u
e

(
d
B
)
Frequency (rad/s)
Sensitivity
Complementary Sensitivity

Fig. 10. Frequency domain indicators of the quarter vehicle MR suspension system
-K
LQG
B
C
z
r

System
Linearized
Model
C
L
u(t)
x
x
y(t)
+
-


Sliding surface
2
x
1
x
) (
0
t x
) (
1
t x
0 = s
Reaching
Phase
Sliding Phase

Fig. 11. Scheme of the sliding mode control
where c
(i = 1, 2, S, 4) are sliding surface coefficients to be determined so that the sliding

surface is stable. Then the sliding mode controller can be formulated which satisfies sliding
mode condition ss < u as follows (Choi et al., 2000):
u =
:
MRD
c
S
[c
1
x
2
+
c
2
m
s
(k
s
x
1
c
s
x
2
+ k
s
x
3
+c
s
x
4
x
5
) +c
3
x
4
+
c
4
m
u
(k
s
x
1
+c
s
x
2

(k
t
+k
s
)x
3
c
s
x
4
+x
5
) +
1
:
MRD
x
5
ksgn(s), (k > u)

(14)
where k stands for the discontinuous gain which is a positive number. The discrete gain can
be changed to continuous gain for reducing the chattering problem.
sat(s) = _
se or |se| 1
sgn(s) or |se| > 1

(15)
3.5 Moving sliding mode controller (MSMC)
It is required to reduce reaching phase to improve the performance and robustness of the
SMC. Reaching time can be successfully reduced in case that sliding surface is rotated or
shifted by considering reaching phase. Fig. 12 illustrates two moving patterns of the sliding
surface: rotating and shifting. The sliding surface for moving sliding mode controller can be
determined as follows:
s = c
1
(x
1
, x
2
)x
1
+c
2
x
2
+c
3
x
3
+ c
4
x
4
+ x
5
+ o(x
1
, x
2
)
c
1
(x
1
, x
2
) =
(A
]r
-x
2
)
x
1
, o(x
1
, x
2
) = c
1o
x
1
+ x
2

]s

(16)
where
]
and
]s
are gain for surface rotating and surface shifting respectively. c
1o
is initial
value of c
1
. The moving sliding mode controller can be formulated as follows:
u =
:
MRD
c
S
_c
1
(x
1
, x
2
)x
2
+
c
2
m
s
(k
s
x
1
c
s
x
2
+ k
s
x
3
+c
s
x
4
x
5
) + c
3
x
4
+
c
4
m
u
(k
s
x
1
+c
s
x
2

(k
t
+k
s
)x
3
c
s
x
4
+x
5
) +
1
:
MRD
x
5
+o(x
1
, x
2
)_ ksgn(s), (k > u)

(17)

3.6 Fuzzy moving sliding mode controller (FMSMC)
The fuzzy moving sliding mode controller, which can change the coefficients and intercepts of
sliding surface by fuzzy tuning which takes into account for location of reaching phase, is
developed. Fig. 13 presents the block diagram of the proposed FMSMC. The basic
configuration of fuzzy control consists of three components: a fuzzification interface, a
decision-making logic and a deffuzification interface. In this study, the coefficient of sprung
mass displacement x
1
is considered with priority while other coefficients of sliding surface are
fixed. Rotating algorithm, which can change the sliding surface c
1
as a function of x
1
, is
applied to reduce the reaching phase. However, it is hard to expect that the robustness is
enhanced in case that initial condition of reaching phase is located in quadrant 1 or quadrant 3.
Hence, shifting algorithm, which can change the intercept o as a function of x
1
, is adopted and
determined according to the relative place of sliding surface, and this will sufficiently reduce
the reaching phase. Therefore, the sliding surface can be written as follows (Cho et al., 2007):
s = c
1
(x
1
)x
1
+c
2
x
2
+c
3
x
3
+ c
4
x
4
+ o(x
1
)

(18)

min
c
max
c
min
c
max
c
1
x
2
x
O
) (
0
t x

(a) rotating sliding surface

(b) shifting sliding surface
Fig. 12. Scheme of the moving sliding mode control


Fig. 13. Block-diagram of the proposed FMSMC
where c
1
and o are determined by tuning of fuzzy logic which takes into account for
displacement and direction of sprung mass displacement of x
1
. In this case, the sliding
surface should maintain stability, although c
1
and o are changed according to time. In order
to reduce reaching time in the rotating algorithm, c
1
should be high in case of the opposite
case. In the letter case, the sprung displacement x
1
can be converted toward 0 more quickly.
In the shifting control algorithm, absolute value of o should be high in case that absolute
value of x
1
is large and vice versa. Furthermore, sliding surface is moved upward for
positive x
1
and sliding surface is moved downward for the opposite case. Consequently, we
can formulate the following control input of the proposed FMSMC, which is combined
rotating and shifting algorithm:
u =
:
MRD
c
S
][c
1
(x
1
) +
Ac
1
1r
sInh (
1r
)
cosh
2
(
1r
)

(u
mcx
c
s
)
cosh
2
(
1s
)
x
2
+
c
2
m
s
(k
s
x
1
c
s
x
2
+k
s
x
3
+c
s
x
4
x
5
) +
c
3
x
4
+
c
4
m
u
(k
s
x
1
+ c
s
x
2
(k
t
+ k
s
)x
3
c
s
x
4
+ x
5
)
1
:
MRD
x
5
ksgn(s), (k > u)

(19)
The control input determined from the FMSMC is to be applied to the MR damper
depending upon the motion of suspension travel. The detailed control algorithm was
described by Cho et al., 2007.
3.7 Semi-active condition
The control input u directly represents the damping force of F
MR
. On the other hand, the
damping of the suspension system needs to be controlled depending upon the motion of
suspension travel. Therefore, the following semi-active actuating condition is imposed:
u = _
u or u(z
s
z
u
) > u
u or u(z
s
z
u
) u

(20)
This semi-active condition can assure the increment of damping characteristic of the
suspension system and hence increase the stability of the system.
4. Control performances
Vibration control performances of the quarter vehicle MR suspension system are evaluated
under two types of excitation (road) conditions. The first excitation, normally used to reveal
the transient response characteristic, is a bump described by
MSMC
c1
Defuzzification
Decision
Making
Fuzzification
Defuzzification
Decision
Making
Fuzzification
(Rotating)
(Shifting)
MR Suspension
x1
x1
Fuzzy Control

z
= A
m
|1 cos (t)]

(21)
where = 2n, = 1I and I = I. A
m
(= u.u7m) is the bump height, (= u.8m) is the
width of the bump and I is the vehicle velocity. In the bump excitation, the vehicle travels
the bump with a constant vehicle velocity of 3.08km/h(=0.856m/s). The second excitation is
a random road conditions.
Fig. 14 shows the experimental apparatus of the quarter vehicle MR suspension system to
evaluate and compare the effectiveness of vibration isolation of the proposed control
algorithms. The MR suspension (assembly of the MR damper and spring), sprung mass and
tire are installed on the hydraulic excitation system. The sprung mass displacement and
excitation displacement are measured by LVDT (linear variable differential transformer) and
the suspension travel is measured by a wire sensor. The hydraulic system applies the road
profile to the MR suspension system. The current amplifier applies the control current
determined from the control algorithm to the MR damper.

Fig. 14. Experimental apparatus of the quarter vehicle MR suspension system
Fig. 15 presents vertical displacement, vertical acceleration of sprung mass and suspension
travel versus time responses of the MR suspension system for the bump excitation. As
shown in the Fig., it can obviously be found that unwanted vibrations induced from the
bump excitation are well suppressed by adopting the control algorithms in the MR
suspension system. Fig. 16 shows the performance comparison of vertical acceleration RMS
of the quarter vehicle MR suspension system for the bump excitation. The RMS value is
calculated from:
RHS =
_
x
1
2
+x
2
2
++x
n
2
n

(22)
As shown in Fig., skyhook control algorithm shows quite good vibration control
performance. Also, Fig. 17 shows the PSD (power spectrum density) of sprung mass
acceleration in frequency domain for the random excitation and Fig. 18 shows the
performance comparison of vertical acceleration RMS of the quarter vehicle MR suspension
system for the random excitation. As expected, the PSD of the vertical acceleration has been
Acc.
Current
Input Signal
Disp.
Computer
AD/DA
Converter
& Amplifier
Accelerometer
MR Suspension
LVDT
Hydraulic Unit
LVDT
Disp.
Wire
Sensor
Disp.

1 2 3 4
-0.05
0.00
0.05
0.10
0.15
Uncontrolled Skyhook
LQG Hinf

V
e
r
t
i
c
a
l

D
i
s
p
.

o
f

S
M

(
m
)
1 2 3 4
-3
-2
-1
0
1
2
3

V
e
r
t
i
c
a
l

A
c
c
.

o
f

S
M

(
m
/
s
2
)
1 2 3 4
-0.03
-0.02
-0.01
0.00
0.01
0.02
0.03

S
u
s
p
e
n
s
i
o
n

T
r
a
v
e
l

(
m
)
Time (sec)
1 2 3 4
-0.05
0.00
0.05
0.10
0.15
Uncontrolled SMC
MSMC FMSMC

V
e
r
t
i
c
a
l

D
i
s
p
.

o
f

S
M

(
m
)
1 2 3 4
-3
-2
-1
0
1
2
3

V
e
r
t
i
c
a
l

A
c
c
.

o
f

S
M

(
m
/
s
2
)
1 2 3 4
-0.03
-0.02
-0.01
0.00
0.01
0.02
0.03

S
u
s
p
e
n
s
i
o
n

T
r
a
v
e
l

(
m
)
Time (sec)

Fig. 15. Bump responses of the quarter vehicle MR suspension system

UC SkyH LQG Hinf SMC MSMC FMSMC
0.0
0.1
0.2
0.3
0.4

R
M
S

o
f

V
e
r
t
i
c
a
l

A
c
c
.

Fig. 16. Performance comparison of vertical acceleration RMS of the quarter vehicle MR
suspension system (bump)

1 10
0
2
4
6
8
10
12
14
16
0.5

P
S
D

o
f

S
M

A
c
c
.

(
(
m
/
s
)
2
/
H
z
)
Frequency (Hz)
Uncontrolled
Skyhook
LQG
Hinf
50

1 10
0
2
4
6
8
10
12
14
16
0.5

P
S
D

o
f

S
M

A
c
c
.

(
(
m
/
s
)
2
/
H
z
)
Frequency (Hz)
Uncontrolled
SMC
MSMC
FMSMC
50

Fig. 17. Random responses of the quarter vehicle MR suspension system (72km/h)

UC SkyH LQG Hinf SMC MSMC FMSMC
0.0
0.5
1.0
1.5

R
M
S

o
f

V
e
r
t
i
c
a
l

A
c
c
.

Fig. 18. Performance comparison of vertical acceleration RMS of the quarter vehicle MR
suspension system (random, 72km/h)

considerably reduced in the neighbourhood of body resonance (1~2Hz) by applying control
input. The control results presented in Figs. 15~18 indicate that ride comfort of a vehicle
system can be substantially improved by employing the MR suspension system associated
with the proposed control algorithms.
5. Conclusion
In this chapter, control algorithms for vibration control of the quarter vehicle MR
suspension system were proposed and its effectiveness was experimentally verified and
compared. In order to achieve this goal, a cylindrical MR damper was designed and
manufactured. After evaluating the field-dependent damping characteristics of the MR
damper, a quarter vehicle suspension system was then constructed and its governing
equations of motion were derived. In order to obtain a favourable control performance of
the MR suspension system, skyhook controller, LQG controller, H controller, sliding mode
controller, moving sliding mode controller and fuzzy moving sliding mode controller were
designed and experimentally realized to the quarter vehicle MR suspension system. It has
been experimentally shown that the proposed control algorithms can reduce the unwanted
vibration under bump excitation. In addition, it has been demonstrated that using the
proposed control methodologies vibration levels such as sprung mass acceleration under
random excitation can be significantly reduced at body resonance region. The results
presented here are self-explanatory justifying that the control strategies implemented in this
work can produce very similar vibration control performance. This directly indicates that
ride comfort of the vehicle system can be substantially improved by adopting the MR
suspension system associated with the proposed control strategies.
6. References
Bahram, S. and Michael, H. (1993). Control System Design Using MATLAB, Prentice Hall,
ISBN 0-13-174061-X, New Jersey
Carlson, J. D., Cantanzarite, D. M. and St.Clair, K. A. (1996). Commercial
Magnetorheological Fluid Devices. Electro-Rheological Fluids, Magneto-Rheological
Suspensions and Associated Technology, Vol.1996, pp. 20-28, ISBN 981-02-2676-4
Cho, J. W., Pang, Y. S., Choi, S. B. and Kim, K. S. (2007). Fuzzy Moving Sliding Mode
Control of Vehicle Suspension Featuring ER Shock Absorber. International
Conference on Electro-Rheological Fluids and Magneto-Rheological Suspensions, ISBN
978-981-2771-19-3, Lake Tahoe, USA
Choi, S. B., Han, S. S., Kim, H. K. and Cheong, C. C. (1999). H Control of a Flexible Gantry
Robot Arm Using Smart Actuators. Mechatronics: Mechanics, Electronics, Control,
Vol.9, No.3, pp. 271-286, ISSN 0957-4158
Choi, S. B., Choi, Y. T. and Park, D. W. (2000). A Sliding Mode Control of a Full-Car
Electrorheological Suspension System Via Hardware in-the-Loop Simulation.
Journal of Dynamic Systems, Measurement, and Control, Vol.122, No.1, pp. 114-121,
ISSN 0022-0434
Choi, S. B., Lee, S. K. and Park, Y. P. (2001). A Hysteresis Model for the Field-Dependent
Damping Force of a Magnetorheological Damper. Journal of Sound and Vibration,
Vol.245, No.2, pp. 375-383, ISSN 0022-460x

Choi, S. B., Lee, H. S. and Park, Y. P. (2002). H Control Performance of a Full-Vehicle
Suspension Featuring Magnetorheological Dampers. Vehicle System Dynamics,
Vol.38, No.5, pp. 341-360, ISSN 0042-3114
Du, H., Yim, S. K. and Lam, J. (2005). Semi-Active H Control of Vehicle Suspension with
Magneto-Rheological Dampers. Journal of Sound and Vibration, Vol.283, No.3/5, pp.
981-996, ISSN 0022-460x
Guo, D. and Hu, H. (2005). Nonlinear Stiffness of a Magneto-Rheological Damper. Nonlinear
Dynamics, Vol.40, No.3, pp. 241-249, ISSN 0924-090x
Hong, S. R., Wereley, N. M., Choi, Y. T. and Choi, S. B. (2008). Analytical and Experimental
Validation of a Nondimensional Bingham Model for Mixed-Mode
Magnetorheological Dampers. Journal of Sound and Vibration, Vol.312, No.3, pp. 399-
417, ISSN 0022-460x
Kamath, G. M., Wereley, N. M. and Jolly, M. R. (1998). Characterization of Semiactive
Magnetorheological Helicopter Lag Mode Dampers. Smart Structures and Integrated
Systems, Vol.1998, pp. 356-377, ISSN 0277-786x
Kim, J. H., Kang, B. W., Park, K. M., Choi, S. B. and Kim, K. S. (2002). MR Inserts For Shock
Wave Reduction in Warship Structures. Journal of Intelligent Material Systems and
Structures, Vol.13, No.10, pp. 661-666, ISSN 1045-389x
Lee, H. G., Sung, K. G., Choi, S. B., Park, M. K. and Park, M. K. (2011). Performance
Evaluation of a Quarter-Vehicle MR Suspension System with Different Tire
Pressure. International Journal of Precision Engineering and Manufacturing, Vol.12,
No.2, pp. 203-210, ISSN 1229-8557
Liu, B., Li, W. H., Kosasih, P. B. and Zhang, X. Z. (2006). Development of an MR-Brake-
Based Haptic Device. Smart Materials and Structures, Vol.15, No.6, pp. 1960-1966,
ISSN 0964-1726
Ok, S. Y., Kim, D. S., Park, K. S. and Koh, H. M. (2007). Semi-Active Fuzzy Control of Cable-
Stayed Bridges Using Magneto-Rheological Dampers. Engineering Structures,
Vol.29, No.5, pp. 776-788, ISSN 0141-0296
Pranoto, T. and Nagaya, K. (2005). Development on 2DOF-Type and Rotary-Type Shock
Absorber Damper Using MRF and Their Efficiencies. Journal of Materials Processing
Technology, Vol.161, No.1/2, pp. 146-150, ISSN 0924-0136
Seong, M. S., Choi, S. B. and Han, Y. M. (2009). Damping Force Control of a Vehicle MR
Damper Using a Preisach Hysteretic Compensator. Smart Materials and Structures,
Vol.18, No.7, pp. 1-13, ISSN 0964-1726
Shen, Y., Golnaraghi, M. F. and Heppler, G. R. (2007). Load-Leveling Suspension System
with a Magnetorheological Damper. Vehicle System Dynamics, Vol.45, No.4, pp. 297-
312, ISSN 0042-3114
Spencer Jr., B. F., Dyke, S. J., Sain, M. K. and Carlson, J. D. (1997). Phenomenological Model
for Magnetorheological Dampers. Journal of Engineering Mechanics, Vol.123, No. 3,
pp. 230-238, ISSN 0733-9399
White, F. M. (2003). Fluid Mechanics (Fifth Edition), Mcgraw-Hill, ISBN 0-07-240217-2, New
York
Yu, M., Liao, C. R., Chen, W. M. and Huang, S. L. (2006). Study on MR Semi-Active
Suspension System and its Road Testing. Journal of Intelligent Material Systems and
Structures, Vol.17, No.8/9, pp. 801-806, ISSN 1045-389x
Yu, M., Dong, X. M., Choi, S. B. and Liao, C. R. (2009). Human Simulated Intelligent Control
of Vehicle Suspension System with MR Dampers. Journal of Sound and Vibration,
Vol.319, No.3/5, pp. 753-767, ISSN 0022-460x
6
A Semiactive Vibration Control Design for
Suspension Systems with Mr Dampers
Hamid Reza Karimi
Department of Engineering, Faculty of Engineering and Science
University of Agder
Norway
1. Introduction
In an automotive system, the vehicle suspension usually contributes to the vehicle's
handling and braking for good active safety and driving pleasure and keeps the vehicle
occupants comfortable and reasonably well isolated from road noise, bumps and vibrations.
The design of vehicle suspension systems is an active research field in automotive industry
(Du and Zhang, 2007; Guglielmino, et al., 2008). Most conventional suspensions use passive
springs to absorb impacts and shock absorbers to control spring motions. The shock
absorbers damp out the motions of a vehicle up and down on its springs, and also damp out
much of the wheel bounce when the unsprung weight of a wheel, hub, axle and sometimes
brakes and differential bounces up and down on the springiness of a tire.
Semiactive suspension techniques (Karkoub and Dhabi, 2006; Shen, et al., 2006; Zapateiro, et
al., 2009) promise a solution to the problem of vibration absorption with some
comparatively better features than active and passive devices. Compared with passive
dampers, active and semiactive devices can be tuned due to their flexible structure. One of
the drawbacks of active dampers is that they may become unstable if the controller fails. On
the contrary, semiactive devices are inherently stable, because they cannot inject energy to
the controlled system, and will act as pure passive dampers in case of control failure.
Among semiactive control devices, magnetorheological (MR) dampers are particularly
interesting because of the high damping force they can produce with low energy
requirements (being possible to operate with batteries), simple mechanical design and low
production costs. The damping force of MR dampers is produced when the MR fluid inside
the device changes its rheological properties in the presence of a magnetic field. In other
words, by varying the magnitude of an external magnetic field, the MR fluid can reversibly
go from a liquid state to a semisolid one or vice versa (Carlson, 1999). Despite the above
advantages, MR dampers have a complex nonlinear behavior that makes modeling and
control a challenging task. In general, MR dampers exhibit a hysteretic force - velocity loop
response whose shape depends on the magnitude of the magnetic field and other variables.
Diverse MR damper models have been developed for describing the nonlinear dynamics
and formulating the semiactive control laws (Dyke, et al., 1998; Zapateiro and Luo, 2007;
Rodriguez, et al., 2009). Most of the MR dampers models found in literature are the so-
called phenomenological models which are based on the mechanical behavior of the device
(Spencer, et al., 1997; Ikhouane and Rodellar, 2007).


116
The objective of the work is to mitigate the vibration in semiactive suspension systems
equipped with a MR damper. Most conventional suspensions use passive devices to absorb
impacts and vibrations, which is generally difficult to adapt to the uncertain circumstances.
Semiactive suspension techniques promise a solution to the above problem with some
comparatively better features than active and passive suspension devices. To this aim, a
backstepping control is proposed to mitigate the vibration in this application. In the design
of backstepping control, the Bouc-Wen model of the MR damper is used to estimate the
damping force of the semiactive device taking the control voltage and velocity inputs as
variables and the semiactive control law takes into account the hysteretic nonlinearity of the
MR damper. The performance of the proposed semiactive suspension strategy is evaluated
through an experimental platform for the semiactive vehicle suspension available in our
laboratory.
The chapter is organized as follows. In the section 2, physical study of MR dampers is
proposed. The mathematical model for the semiactive suspension experimental platform is
introduced in the section 3. In the section 4, details on the formulation of the backstepping
control are given. The results of control performance verification are presented and
discussed in the section 5. Finally, conclusions are drawn at the end of the paper.
2. MR damper
Nowadays dampers based on MagnetoRheological (MR) fluids are receiving significant
attention especially for control of structural vibration and automotive suspension systems. .
In most cases it is necessary to develop an appropriate control strategy which is practically
implementable when a suitable model of MR damper is available. It is not a trivial task to
model the dynamic of MR damper because of their inherent nonlinear and hysteretic
dynamics. In this work, an alternative representation of the MR damper in term of neural
network is developed. Training and validating of the network models are achieved by using
data generated from the numerical simulation of the nonlinear differential equations
proposed for MR damper. The MR damper is a controllable fluid damper which belongs in
the semi-active category. A brief overview of the physical buildup of an MR damper is seen
in this section.
2.1 Physical study
The MR damper has a physical structure much like a typical passive damper: an outer
casing, piston, piston rod and damping fluid confined within the outer casing. The main
difference lies in the use of MR fluid and an electromagnet.
2.1.1 MR fluid
A magneto rheological fluid is usually a type of mineral or silicone oil that carries magnetic
particles. These magnetic particles may be iron particles that can measure 3-10 microns in
diameter, shown in Fig. 1. In addition to these particles it might also contain additives to
keep the iron particles suspended. When this fluid is subject to a magnetic field the iron
particles behave like dipoles and start aligning along the constant flux, shown in Fig.
2.When the fluid is contained between the dipoles, its movement is restricted by the chain of
the particles thus increasing its viscosity. Thus it changes its state from liquid to a
viscoelastic solid.

A Semiactive Vibration Control Design for Suspension Systems with Mr Dampers

117

Fig. 1. Magnetic particles in the MR fluid.

Fig. 2. Particles aligning along the flux lines.
Mechanical properties of the fluid in its on state are anisotropic i.e. it is directly dependent
on the direction. Hence while designing a MR device it is important to ensure that the lines
of flux are perpendicular to the direction of the motion to be restricted. This way the yield
stress of the fluid can be controlled very accurately by varying the magnetic field intensity.
Controlling the yield stress of a MR fluid is important because once the peek of the yield
stress is reached the fluid cannot be further magnetized and it can result in shearing. It is
also known that the MR Fluids can operate at temperatures ranging from -40 to 150 C with
only slight changes in the yield stress. Hence it is possible to control the fluids ability to
transmit force with an electromagnet and make use of it in control-based applications.
2.1.2 Electromagnet
The electromagnet in the MR damper can be made with coils wound around the piston. An
example is the MR damper design by Gavin et. al (2001), seen in Fig. 3. The wire connecting
this electromagnet is then lead out through the piston shaft.
2.2 Modes of operation
MR Fluids can be used in three different modes (Spencer et al, 1997):
Flow mode: Fluid is flowing as a result of pressure gradient between two stationary plates. It
can be used in dampers and shock absorbers, by using the movement to be controlled to
force the fluid through channels, across which a magnetic field is applied, see Fig. 4.


118
Shear mode: In this mode the fluid is between two plates moving relative to one another. It
is used in clutches and brakes i.e. in places where rotational motion must be controlled,
see Fig. 5.

Fig. 3. Electromagnetic piston.

Fig. 4. Flow mode.

Fig. 5. Shear mode.
Squeeze-flow mode: In this mode the fluid is between two plates moving in the direction
perpendicular to their planes. It is most useful for controlling small movements with large
forces, see Fig. 6.


119

Fig. 6. Squeeze flow mode.
2.3 MR damper categories
2.3.1 Linear MR dampers
There are three main types of linear MR dampers, the mono, twin and double-ended MR
dampers (Ashfak et. al, 2011). All of these have the same physical structure of an outer
casing, piston rod, piston, electromagnet and the MR fluid itself.
2.3.2 Mono and twin
The mono damper is named because of its single MR fluid reservoir. As the piston displaces
due to an applied force, the MR liquid compresses the gas in the gas reservoir. Just like the
other two MR damper types, the mono MR damper has its electromagnets located in the
piston. Fig. 7 shows a schematic diagram of the mono MR damper.
The twin MR damper has two housings, see Fig. 8. Other than this, it is identical to the mono
MR damper.

Fig. 7. The mono MR damper.

Fig. 8. The twin MR damper.


120
2.3.3 Double-ended
The double-ended MR damper is named so because of the double protruding pistons from
both ends of the piston, see Fig. 9. No gas accumulators are used in this setup because the
MR fluid is able to squeeze from one chamber to the other. In an experimental design by
Lord Corp, a thermal expansive accumulator is used. This is to store the expanded liquid
due to heat generation, see Fig. 10.

Fig. 9. Double-ended MR damper.

Fig. 10. Double-ended MR damper with thermal expansion accumulator.
2.3.4 Rotary dampers
Rotary dampers, as the name suggests, are used when rotary motion needs damping. There
exist several types of rotary dampers, but the one that will be described is the disk brake.
This is also the type that is used on the SAS platform.
The disk brake is one of the most commonly used rotary dampers. It has a disk shape and
contains MR fluid and a coil as shown in Fig. 11. Different setups have been proposed for
the MR disk brake. A comparison of these has been done by Wang et al (2004) and Carlson
et al (1998).


121

Fig. 11. MR brake disk.
3. Problem formulation
The experimental platform used in this work is fabricated by the Polish company Inteco
Limited, see Fig. 12. It consists of a rocking lever that emulates the car body, a spring, and an
MR damper that makes the semiactive vibration control. A DC motor coupled to an
eccentric wheel is used to simulate the vibrations induced to the vehicle. Thus, the higher is
the motor angular velocity, the higher is the frequency of the car (rocking lever) vibrations.
The detailed definitions of the angles and distances

can be found in the appendix.

Fig. 12. Picture of the SAS system (Inteco Ltd., Poland).
The equations of motion of the upper rocking lever are given by:

( )
( )
2 2
1 1
2 2 21 22 2 2 2 2 2
sin
s f eq mr f f
J M M M J r F f

=
= + + +
(1)


122
where
2
and
2
are the angular position and angular speed of the upper lever, respectively.
M
21
, M
22
and M
s2
are the viscous friction damping torque, the gravitational forces torque and
the spring torque acting on the lower rocking lever, respectively and their equations are:

( )
( )
21 2 2
22 2 2 2
2 2 2 2
cos
sin
s s s s s
M k
M G R
M r F

=
=
= +
(2)
F
s
is the force generated by the spring and
s
is the slope angle of the spring operational line,
which are given by:
( ) ( )
2
0 2 2 2 1 1 1 1 1 1 2 2 2
sin( ) sin( ) cos( ) cos( )
s s s s s s s s s s s
F K l r r b r r

= + +

(3)

1 1 1 1 2 2 2
1 1 1 2 2 2
sin( ) sin( )
tan
cos( ) cos( )
s s s s
s
s s s s
r r
abs
b r r

=

(4)
F
eq
f
mr
is the force generated by the MR damper:

2 2 2 2 1 1 1 1
( cos cos )
2 2
eq mr mr f f f f f f
F f f r r

= + + + + + + +

(5)

1 1 1 2 2 2
1
1 1 1 2 2 2
sin( ) sin( )
tan
cos( ) cos( )
f f f f
f
f f f f
r r
abs
b r r

+ +
=
+ +

(6)
The model is completed with the equations of motion of the lower rocking lever:

( )
1 1
1
1 1 11 12 13 14 1 1 s f
J M M M M M M

=
= + + + + +
(7)
with

( ) ( ) ( ) ( )
( )
( ) ( )
( ) ( )
11 1 1
12 2 2 2
13 1 1 0 1 1
14 1 1
1 1 1 1
1 1 1 1
cos( )
cos sin ( )
( )
cos
sin
sin
g x
g
s s s s s
f f f f f
M k
M G R
M R K l r R D e t
de t
M f R
dt
M r F
M r F

=
=
= + + + + +

= +

=
= +
(8)
where M
11
is the viscous friction damping torque; M
12
is the gravitational forces torque; M
13

is the actuating kinematic torque transferred through the tire; M
14
is the damping torque
generated by the gum of tire; M
s1
is the torque generated by the spring; M
f1
is the torque
generated by the damper, and e(t) is the disturbance input.
The objective of the semiactive suspension is to reduce the vibrations of the car body (the
upper rocking lever). This can be achieved by reducing the angular velocity of the lever
2
.


123
Thus, the system to be controlled is that of (1) by assuming that the lower rocking lever
dynamics constitute the disturbances.
4. Backstepping control design
For making the backstepping control design, define z
1
and z
2
as the new coordinates
according to:

( )
( )
1 2 2 2 2
, ,
equ
z z =
(9)
where the equilibrium point of the system is
( )
2 2
,
equ equ
= ( ) 0.55 , 0 , 0
mr
rad f = . The
above change of coordinates is made so that the equilibrium point is set to (0, 0). In the new
coordinates, (1) becomes:

( )
( )
1 2
1 1
2 2 21 22 2 2 2 2 1 2
sin
s f eq f equ f mr mr
z z
z J M M M J r F z f f g f

=
= + + + = +
(10)
The backstepping technique can now be applied to the system (10). First, define the
following standard backstepping variables and their derivatives:

1 1 1 2
2 2 1 2 2 1 2
1 1 1 1 1 1 2
, 0
e z e z
e z e z h z
h e h h z

= =
= = +
= > =
(11)
For the control design, the following Bouc-Wen model of the MR damper (Ikhouane and
Dyke, 2007) is used:
( )
mr
f v w c(v)x = + (12a)

n n
w x w w x w x = + (12b)
( )
0 1
c v c c v = + (12c)
( )
0 1
v v = + (12d)
where v is the control voltage and w is a variable that accounts for the hysteretic dynamics.
, c, , , n, are parameters that control the shape of the hysteresis loop. From control
design point of view, it is desirable to count on the inverse model, i.e., a model that predicts
the control voltage for producing the damping force required to reduce the vibrations. This
is because the force cannot be commanded directly; instead, voltage or current signals are
used as the control input to approximately generate the desired damping force.
Now, define the following Lyapunov function candidate:

1 2
2 2 2 2
1 2
1 1 1 1
2 2 2 2
V V V e e = + = + (13)


124
Deriving (13) and substituting (10)-(11) in the result yields:

2
1 1 2 2 1 2 1 1 2 2 1 2 2
2 2
1 1 2 2 2 2 2 1 2 1 2 2
( )(1 ) ( )
mr
equ mr
V e e e e e e h e e f e g f h z e
h e h e e h h h h f g f
= + = + + +

= + + + + + +

(14)
In order to make ) t ( V
negative, the following control law is proposed to generate the

force
mr
f :

( )
( ) ( )
2 2 1 2 1 2 2
1

equ
mr
h h h h f
f
g
+ + + +
= (15)
Substitution of (15) into (14) yields:

2 2
1 1 2 2 1 2
0, , 0 V h e h e h h = < >

(16)
Thus, according to the Lyapunov stability theory, the system is asymptotically stable.
Therefore,
1
0 e and
2
0 e , and consequently
2 2equ
and
2
0 by using the
control law (15).
Note that the control force f
mr
in (15) cannot be commanded directly, thus voltage or current
commanding signals are used as the control input to approximately generate the desired
damping force. Concretely, by making use of the Dahl model (12), the following voltage
commanding signal is obtained from (15):

( )
( ) ( ) ( )
2 2 1 2 1 2 2 0 0
1 1
1
0
( )
( )
0 otherwise
equ
h h h h f g c x w
g
v t
w c x g

+ + + + +

=
+
(17)
which is the control signal that can be sent to the MR damper.
5. Simulation results
In this section, MR damper parameters
0
= 1,8079,
1
= 8,0802, c
0
= 0,0055, c
1
= 0,0055, =
84,0253, = 100, n = 1 and = 80,7337 (Ikhouane and Dyke, 2007) are taken for the simulation.
The displacement curves and velocity curves showing hysteresis of the three last simulations,
with different values of voltage, are given in Fig. 13 and Fig. 14, respectively. The blue curve is
for no current, and gives the effect of the passive damper. We notice that the higher current the
higher torque and less hysteresis width. All of the curves starts wide, and gets smaller and
closer to zero by time. This is because of the damping. The system is stable.
Now, the backstepping control law (17) was applied to the experimental platform with the
parameters 1 h
1
= and 10 h
2
= for the simulation.
The effectiveness of the backstepping controller for the vibration reduction can be seen in
Fig. 15. It shows the system response (angular position and velocity) for three different
excitation inputs: step, pulse train and random excitation. The figures show the comparison
of the system response in two cases: no control, when the current to MR damper is 0 A at
all times (or equivalently, the voltage is set to 0 V) and Backstepping, when the controller


125
is activated. The reduction in the RMS angular velocity achieved in each case is 43.5%, 37.3%
and 40.7%, respectively.

-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
displacement (cm)
f
o
r
c
e

(
N
)
v=0.2
v=0.1
v=0

Fig. 13. Displacement vs torque.

-5 -4 -3 -2 -1 0 1 2 3 4 5
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
velocity (cm/s)
f
o
r
c
e

(
N
)
v=0.2
v=0.1
v=0

Fig. 14. Velocity vs torque.


126

(a)

(b)


127

(c)

Fig. 15. Suspension systems response with the backstepping control: (a) Step input; (b) Pulse
train input; (c) Random input.
6. Conclusions
In this paper we have studied the application of semiactive suspension for the vibration
reduction in a class of automotive systems by using MR dampers. Backstepping and
heuristic controllers have been proposed: the first one is able to account for the MR
dampers nonlinearities and the second one needs only the information of the measured
vibration. The control performance has been evaluated through the simulations on an
experimental vehicle semiactive suspension platform. It has been shown that the proposed
semiactive control strategies are capable of reducing the suspension deflection with a
significantly enhanced control performance than the passive suspension system.


128
7. Appendix. Geometrical diagram (Inteco SAS manual)

Geometrical diagram (Inteco SAS Manual)
where
r
1
= r
2
= 0.025 m: distance between the spring joint and the lower and upper rocking
lever line.
r
3
= 0.050 m: distance between the wheel axis and the lower rocking lever line.
l
1
= 0.125 m: distance between the damper joint and the lower rocking lever line.
l
2
= 0.130 m: distance between the damper joint and the upper rocking lever line.
l
3
= 0.200 m: distance between the wheel axis and the lower rocking lever line.
s
1
= 0.135 m: distance between the spring joint and the lower rocking lever line.
s
2
= 0.160 m: distance between the spring joint and the upper rocking lever line.


129

1f
= 0.2730 rad: damper fixation angle.

2f
= 0.2630 rad: damper fixation angle.

1s
= 0.1831 rad: spring fixation angle.

2s
= 0.1550 rad: spring fixation angle.
= 0.2450 rad: wheel axis fixation angle.
r
1f
= 0.1298 m: lower rotational radius of the damper suspension.
r
2f
= 0.1346 m: upper rotational radius of the damper suspension.
r
1s
= 0.1373 m: lower rotational radius of the spring suspension.
r
2s
= 0.1619 m: upper rotational radius of the spring suspension.
R = 0.2062 m: rotational radius of the wheel axis.
D
x
= 0.249 m: distance between the rocking lever rota-tional axis and the wheel bottom
(minimal eccentricity).
r = 0.06 m: radius of rim.
l
0
= 0.07 m: tire thickness.
b = 0.330 m: distance between the rocking lever rotational axis and car body.
8. References
A. Ashfak, A. Saheed, K. K. Abdul Rasheed, and J. Abdul Jaleel (2011). Design, Fabrication
and Evaluation of MR Damper. International Journal of Aerospace and Mechanical
Engineering. 1, vol. 5, pp. 27-33.
J.D. Carlson (1999), Magnetorheological fluid actuators, in Adaptronics and Smart Structures.
Basics, Materials, Design and Applications, edited by H. Janocha, London: Springer.
J.D. Carlson, D.F. LeRoy, J.C. Holzheimer, D.R. Prindle, and R.H. Marjoram. Controllable
brake. US patent 5,842,547, 1998.
H. Du and N. Zhang (2007), H
control of active vehicle suspensions with actuator time

delay, J. Sound and Vibration, vol. 301, pp. 236-252.
S.J. Dyke, B.F. Spencer Jr., M.K. Sain and J.D. Carlson (1998), An experimental study of MR
dampers for seismic protection, Smart Materials and Structures, vol. 7, pp. 693-703.
H. Gavin, J. Hoagg and M. Dobossy (2001). Optimal Design of MR Dampers. Proc. U.S.-Japan
Workshop on Smart Structures for Improved Seismic Performance in Urban Regions,
Seattle, WA, 2001. pp. 225-236.
E. Guglielmino, T. Sireteanu, C.W. Stammers and G. Ghita (2008), Semi-active Suspension
Control: Improved Vehicle Ride and Road Friendliness, London: Springer.
F. Ikhouane and J. Rodellar (2007), Systems with hysteresis: Analysis, Identification and Control
Using the Bouc-Wen Model, West Sussex: John Wiley & Sons.
F. Ikhouane and S.J. Dyke (2007), Modeling and identification of a shear mode
magnetorheological damper, Smart Materials and Structures, vol. 16, pp. 1-12.
INTECO Limited (2007), Semiactive Suspension System (SAS): Users Manual.
A. Karkoub and A. Dhabi (2006), Active/semiactive suspension control using
magnetorheological actuators, Int. J. Systems Science, vol. 37, pp. 35-44.
A. Rodriguez, F. Ikhouane, J. Rodellar and N. Luo (2009), Modeling and identification of a
small-scale magneto-rheological damper, J. Intelligent Material Systems and
Structures, vol. 20, pp. 825-835.


130
Y. Shen, M.F. Golnaraghi and G.R. Heppler (2006), Semiactive vibration control schemes for
suspension system using magnetorheological damper, J. Vibration and Control, vol.
12, pp. 3-24.
B.F. Spencer, S.J. Dyke, M.K. Sain and J.D. Carlson (1997), Phenomenological model of a
magnetorheological damper, ASCE J. Engineering Mechanics, vol. 123, pp. 230-238.
H. Wang, X.L. Gong, Y.S. Zhu, and P.Q. Zhang (2004). A route to design rotary
magnetorheological dampers. In Proceedings of the Ninth International Conference
on Electrorheological Fluids and Magnetorheological Suspensions, pages 680686,
Beijing, China.
M. Zapateiro and N. Luo (2007), Parametric and non-parametric characterization of a shear
mode MR damper, J. Vibroengineering, vol. 9, pp. 14-18.
M. Zapateiro, N. Luo, H.R. Karimi and J. Vehi (2009), Vibration control of a class of
semiactive suspension system using neural network and backstepping techniques,
Mechanical Systems and Signal Processing, vol. 23, pp. 1946-1953.
0
Control of Nonlinear Active Vehicle Suspension
Systems Using Disturbance Observers
Francisco Beltran-Carbajal
1
, Esteban Chavez-Conde
2
,
Gerardo Silva Navarro
3
, Benjamin Vazquez Gonzalez
1
and Antonio Favela Contreras
4
1
Universidad Autonoma Metropolitana, Plantel Azcapotzalco,
Departamento de Energia, Mexico, D.F.
2
Universidad del Papaloapan, Campus Loma Bonita, Departamento de Ingenieria en
Mecatronica, Instituto de Agroingenieria, Loma Bonita, Oaxaca
3
Centro de Investigacion y de Estudios Avanzados del I.P.N., Departamento de Ingenieria
Electrica, Seccion de Mecatronica, Mexico, D.F.
4
ITESM Campus Monterrey, Monterrey, N.L.
Mexico
1. Introduction
The main control objectives of active vehicle suspension systems are to improve the ride
comfort and handling performance of the vehicle by adding degrees of freedom to the
passive system and/or controlling actuator forces depending on feedback and feedforward
information of the system obtained from sensors.
Passenger comfort is provided by isolating the passengers from the undesirable vibrations
induced by irregular road disturbances and its performance is evaluated by the level of
acceleration by which vehicle passengers are exposed. Handling performance is achieved
by maintaining a good contact between the tire and the road to provide guidance along the
track.
The topic of active vehicle suspension control system, for linear and nonlinear models, in
general, has been quite challenging over the years and we refer the reader to some of the
fundamental works in the vibration control area (Ahmadian, 2001). Some active control
schemes are based on neural networks, genetic algorithms, fuzzy logic, sliding modes,
H-innity, adaptive control, disturbance observers, LQR, backstepping control techniques,
etc. See, e.g., (Cao et al., 2008); (Isermann & Munchhof, 2011); (Martins et al., 2006); (Tahboub,
2005); (Chen & Huang, 2005) and references therein. In addition, some interesting semiactive
vibration control schemes, based on Electro-Rheological (ER) and Magneto-Rheological (MR)
dampers, have been proposedand implemented on commercial vehicles. See, e.g., (Choi et al.,
2003); (Yao et al., 2002).
In this chapter is proposed a robust control scheme, based on the real-time estimation of
perturbation signals, for active nonlinear or linear vehicle suspension systems subject to
unknown exogenous disturbances due to irregular road surfaces. Our approach differs
7
2 Vibration Control
from others in that, the control design problem is formulated as a bounded disturbance
signal processing problem, which is quite interesting because one can take advantage of the
industrial embedded system technologies to implement the resulting active vibration control
strategies. In fact, there exist successful implementations of automotive active control systems
based on embedded systems, and this novel tendency is growing very fast in the automotive
industry. See, e.g., (Shoukry et al., 2010); (Basterretxea et al., 2010); (Ventura et al., 2008);
(Gysen et al., 2008) and references therein.
In our control design approach is assumed that the nonlinear effects, parameter variations,
exogenous disturbances and possibly input unmodeled dynamics are lumped into an
unknown bounded time-varying disturbance input signal affecting a so-called differentially
at linear simplied dynamic mathematical model of the suspension system. The lumped
disturbance signal and some time derivatives of the at output are estimated by using a
at output-based linear high-gain dynamic observer. The proposed observer-control design
methodology considers that, the perturbation signal can be locally approximated by a family
of Taylor polynomials. Two active vibration controllers are proposed for hydraulic or
electromagnetic suspension systems, which only require position measurements.
Some numerical simulation results are provided to show the efciency, effectiveness and
robust performance of the feedforward and feedback linearization control scheme proposed
for a nonlinear quarter-vehicle active suspension system.
This chapter is organized as follows: Section 2 presents the nonlinear mathematical model
of an active nonlinear quarter-vehicle suspension system. Section 3 presents the proposed
vehicle suspension control scheme based on differential atness. Section 4 presents the
main results of this chapter as an alternative solution to the vibration attenuation problem
in nonlinear and linear active vehicle suspension systems actuated electromagnetically or
hydraulically. Computer simulation results of the proposed design methodology are included
in Section 5. Finally, Section 6 contains the conclusions and suggestions for further research.
2. A quarter-vehicle active suspension system model
Consider the well-known nonlinear quarter-vehicle suspension system shown in Fig. 1. In
this model, the sprung mass m
s
denotes the time-varying mass of the vehicle-body and the
unsprung mass m
u
represents the assembly of the axle and wheel. The tire is modeled as a
linear spring with equivalent stiffness coefcient k
t
linked to the road and negligible damping
coefcient. The vehicle suspension, located between m
s
and m
u
, is modeled by a damper and
spring, whose nonlinear damping and stiffness force functions are given by
F
k
(z) = kz + k
n
z
3
F
c
( z) = c z + c
n
z
2
sgn( z)
The generalized coordinates are the displacements of both masses, z
s
and z
u
, respectively. In
addition, u = F
A
denotes the (force) control input, which is applied between the two masses
by means of an actuator, and z
r
(t) represents a bounded exogenous perturbation signal due
Control of Nonlinear Active Vehicle Suspension Systems Using Disturbance Observers 3
Fig. 1. Schematic diagram of a quarter-vehicle suspension system: (a) passive suspension
system, (b) electromagnetic active suspension system and (c) hydraulic active suspension
system.
to irregular road surfaces satisfying:
z
r
(t)
=
1
z
r
(t)
=
2
z
r
(t)
=
3
where
1
= sup
t[0,)
|z
r
(t)|
2
= sup
t[0,)
| z
r
(t)|
3
= sup
t[0,)
| z
r
(t)|
For an electromagnetic active suspension system, the damper is replaced by an
electromagnetic actuator (Martins et al., 2006). In this conguration, it is assumed that
F
c
( z) 0.
The mathematical model of the two degree-of-freedom suspension system is then described
by the following two coupled nonlinear differential equations:
m
s
z
s
+ F
sc
+ F
sk
= u
m
u
z
u
+ k
t
(z
u
z
r
) F
sc
F
sk
= u
(1)
133 Control of Nonlinear Active Vehicle Suspension Systems Using Disturbance Observers
4 Vibration Control
with
F
sk
(z
s
, z
u
) = k
s
(z
s
z
u
) + k
ns
(z
s
z
u
)
3
F
sc
( z
s
, z
u
) = c
s
( z
s
z
u
) + c
ns
( z
s
z
u
)
2
sgn( z
s
z
u
)
where sgn() denotes the standard signum function.
Dening the state variables as x
1
= z
s
, x
2
= z
s
, x
3
= z
u
and x
4
= z
u
, one obtains the following
state-space description:
x
1
= x
2
x
2
=
1
m
s
(F
sc
+F
sk
) +
1
m
s
u
x
3
= x
4
x
4
=
k
t
m
u
x
3
+
1
m
u
(F
sc
+ F
sk
)
1
m
u
u +
k
t
m
u
z
r
(2)
with
F
sk
(x
1
, x
3
) = k
s
(x
1
x
3
) + k
ns
(x
1
x
3
)
3
F
sc
(x
2
, x
4
) = c
s
(x
2
x
4
) + c
ns
(x
2
x
4
)
2
sgn(x
2
x
4
)
It is easy to verify that the nonlinear vehicle suspension system (2) is completely controllable
and observable and, therefore, is differentially at and constructible. For more details
on this topics we refer to (Fliess et al., 1993) and the book by (Sira-Ramirez & Agrawal,
2004). Both properties can be used extensively during the synthesis of different controllers
based on differential atness, trajectory planning, disturbance and state reconstruction,
parameter identication, Generalized PI (GPI) and sliding mode control, etc. See, e.g.,
(Beltran-Carbajal et al., 2010a); (Beltran-Carbajal et al., 2010b); (Chavez-Conde et al., 2009a);
(Chavez-Conde et al., 2009b).
In what follows, a feedforward and feedback linearization active vibration controller, as
well as a disturbance observer, will be designed taking advantage of the differential atness
property exhibited by the vehicle suspension system.
3. Differential atness-based control
The system (2) is differentially at, with a at output given by
L = m
s
x
1
+ m
u
x
3
which is constructed as a linear combination of the displacements of the sprung mass x
1
and
the unsprung mass x
3
.
Then, all the state variables and the control input can be parameterized in terms of the at
output L and a nite number of its time derivatives (Sira-Ramirez & Agrawal, 2004). As a
matter of fact, from L and its time derivatives up to fourth order one can obtain:
L = m
s
x
1
+ m
u
x
3
L = m
s
x
2
+ m
u
x
4
L = k
t
(z
r
x
3
)
L
(3)
= k
t
( z
r
x
4
)
L
(4)
=
1
m
u
u +
k
t
m
u
x
3

1
m
u
(F
sc
+ F
sk
)
k
t
m
u
z
r
+ k
t
z
r
(3)
Therefore, the differential parameterization of the state variables and the control input in the
vehicle dynamics (2) results as follows
x
1
=
m
u
k
t
m
s
L +
1
m
s
L
m
u
m
s
z
r
x
2
=
m
u
k
t
m
s
L
(3)
+
1
m
s
L
m
u
m
s
z
r
x
3
=
1
k
t
L + z
r
x
4
=
1
k
t
L
(3)
+ z
r
u =
1
b
L
(4)
+ a
3
L
(3)
+ a
2
L + a
1
L + a
0
L (t)
(4)
with
a
0
=
k
s
k
t
m
s
m
u
a
1
=
c
s
k
t
m
s
m
u
a
2
=
k
s
m
s
+
k
s
+k
t
m
u
a
3
=
c
s
m
s
+
c
s
m
u
b =
k
t
m
u
and
(t) =
k
ns
k
t
m
u
(x
1
x
3
)
3
c
ns
k
t
m
u
(x
2
x
4
)
2
sgn(x
2
x
4
)
+ k
t
z
r
+
k
t
m
s
+
k
t
m
u
c
s
z
r
+
k
t
m
s
+
k
t
m
u
k
s
z
r
Now, note that from the last equation in the differential parameterization (4), one can see that
the at output satises the following perturbed input-output differential equation:
L
(4)
+ a
3
L
(3)
+ a
2
L + a
1
L + a
0
L = bu + (t) (5)
Then, the at output dynamics can be described by the following 4th order perturbed linear
system:

1
=
2

2
=
3

3
=
4

4
= a
0
1
a
1
2
a
2
3
a
3
4
+ bu + (t)
y =
1
= L
(6)
To formulate the vibration control problem, let us assume, by the moment, a perfect
knowledge of the perturbation term , as well as the time derivatives of the at output up
to third order. Then, from (6) one obtains the following differential atness-based controller:
u =
1
b
+
1
b
(a
3
4
+ a
2
3
+ a
1
2
+ a
0
1
(t)) (7)
with
=
3
4

2
3

1
2

0
1
The use of this controller yields the following closed-loop dynamics:
L
(4)
+
3
L
(3)
+
2
L +
1
L +
0
L = 0 (8)
6 Vibration Control
The closed-loop characteristic polynomial is then given by
p (s) = s
4
+
3
s
3
+
2
s
2
+
1
s +
0
(9)
Therefore, by selecting the design parameters
i
, i = 0, , 3, such that the associated
characteristic polynomial for (8) be Hurwitz, one can guarantee that the at output dynamics
be globally asymptotically stable, i.e.,
lim
t
L (t) = 0
Now, the following Hurwitz polynomial is proposedto get the corresponding controller gains:
p
c
(s) =
s
2
+2
c
c
s +
2
c
2
(10)
where
c
> 0 and
c
> 0 are the natural frequency and damping ratio of the desired
closed-loop dynamics, respectively.
Equating term by term the coefcients of both polynomials (9) and (10 ), one obtains that
0
=
4
c
1
= 4
3
c
2
= 4
2
c
2
c
+2
2
c
3
= 4
c
c
On the other hand, it is easy to show that the closed-loop system (2)-(7) is L
-stable or
bounded-input-bounded-state, that is,
x
1
=
m
u
m
s
1
x
2
=
m
u
m
s
2
x
3
=
1
x
4
=
2
u
k
ns
3
1
2
+ c
ns
2
2
+ c
s
2
+ k
s
+ m
u
3
where =
m
u
m
s
+1.
It is evident, however, that the controller (8) requires the perfect knowledge of the exogenous
perturbation signal z
r
and its time derivatives up to second order, revealing several
disadvantages with respect to other control schemes. Nevertheless, one can take advantage of
the design methodology of robust observers with respect to unmodeled perturbation inputs,
of the polynomial type affecting the observed plant, proposed by (Sira-Ramirez et al., 2008b).
The proposeddisturbance observer is called Generalized Proportional Integral (GPI) observer,
because its design approach is the dual counterpart of the so-called GPI controllers
(Fliess et al., 2002) and whose robust performance, with respect to unknown perturbation
inputs, nonlinear and linear unmodeled dynamics and parametric uncertainties, have been
evaluated extensively through experiments for trajectory tracking tasks on a vibrating
mechanical system by (Sira-Ramirez et al., 2008a) and on a dc motor by (Sira-Ramirez et al.,
2009).
4. Disturbance observer design
In the observer design process it is assumed that the perturbation input signal (t) can be
locally approximated by a family of Taylor polynomials of (r 1)th degree:
(t) =
r1
i=0
p
i
t
i
(11)
where all the coefcients p
i
are completely unknown.
The perturbation signal could then be locally described by the following state-space based
linear mathematical model:
1
=
2
2
=
3
.
.
.
r1
=
r
r
= 0
(12)
where
1
= ,
2
=

,
3
=

, ,
r
=
(r1)
.
An extended approximate state model for the perturbed at output dynamics is then given by

1
=
2

2
=
3

3
=
4

4
= a
0
1
a
1
2
a
2
3
a
3
4
+
1
+ bu
1
=
2
2
=
3
.
.
.
r1
=
r
r
= 0
y =
1
= L
(13)
A Luenberger observer for the system (13) is given by

1
=
2
+
r+3
(y y)

2
=
3
+
r+2
(y y)

3
=
4
+
r+1
(y y)

4
= a
0

1
a
1

2
a
2

3
a
3

4
+

1
+ bu +
r
(y y)
1
=

2
+
r1
(y y)
2
=

3
+
r2
(y y)
.
.
.
r1
=

r
+
1
(y y)
r
=
0
(y y)
y =
1
(14)
8 Vibration Control
The dynamical system describing the state estimation error is readily obtained by subtracting
the observer dynamics (14) from the extended linear system dynamics (6). One then obtains,
with e
1
= y y and e
zi
=
i

i
, i = 1, 2, , r, that
e
1
=
r+3
e
1
+ e
2
e
2
=
r+2
e
1
+ e
3
e
3
=
r+1
e
1
+ e
4
e
4
= (
r
+ a
0
) e
1
a
1
e
2
a
2
e
3
a
3
e
4
+ e
z
1
e
z
1
=
r1
e
1
+ e
z
2
e
z
2
=
r2
e
1
+ e
z
3
.
.
.
e
z
r1
=
1
e
1
+ e
z
r
e
z
r
=
0
e
1
(15)
From this expression, it is not difcult to see that the dynamics of output observation error
e
1
= y y satises the following differential equation:
e
(r+4)
1
+ (
r+3
+ a
3
) e
(r+3)
1
+ (
r+2
+ a
2
+
r+3
a
3
) e
(r+2)
1
+ (
r+1
+ a
1
+
r+2
a
3
+
r+3
a
2
) e
(r+1)
1
+ (
r
+ a
0
+
r+1
a
3
+
r+2
a
2
+
r+3
a
1
) e
(r)
1
+
r1
e
(r1)
1
+ +
2
e
1
+
1
e
1
+
0
e
1
= 0
(16)
which is completely independent of any coefcients p
i
, i = 0, , r 1, of the Taylor
polynomial expansion of (t). This means that, the high-gain observer continuously
self-updates. Therefore, as time goes on, the bounded perturbation input signal (t) is
approximated in the form of a (r 1)th degree time polynomial.
Clearly, the coefcients of the associated characteristic polynomial for (16) can be adjusted, by
means of a suitable specication of the design gains {
r+3
, . . . ,
1
,
0
}, sufciently far from
the imaginary axis in the left half of the complex plane, so that the output estimation error e
1
exponentially asymptotically converges to zero.
A fth-order local mathematical model for the real-time estimation of the perturbation input
signal is proposed in this chapter. Then, the characteristic polynomial for the dynamics of
output observation error is simply given by
p
o1
(s) = s
9
+ (
8
+ a
3
) s
8
+ (
7
+ a
2
+
8
a
3
) s
7
+ (
6
+ a
1
+
7
a
3
+
8
a
2
) s
6
+ (
5
+ a
0
+
6
a
3
+
7
a
2
+
8
a
1
) s
5
+
4
s
4
+
3
s
3
+
2
s
2
+
1
s +
0
(17)
Equating the coefcients of the characteristic polynomial (17) with the corresponding ones of
the following ninth-order Hurwitz polynomial:
p
do1
(s) = (s + p
1
)
s
2
+2
1
1
s +
2
1
4
(18)
one gets the observer gains as follows
0
= p
1
8
1
1
=
8
1
+8p
1
7
1
2
= 8
7
1
1
+24p
1
6
1
2
1
+4p
1
6
1
3
= 24
6
1
2
1
+4
6
1
+32p
1
5
1
3
1
+24p
1
5
1
4
= 32
5
1
3
1
+24
5
1
1
+16p
1
4
1
4
1
+48p
1
4
1
2
1
+6p
1
4
1
5
= 16
4
1
4
1
+48
4
1
2
1
+6
4
1
+32p
1
3
1
3
1
+24p
1
3
1
1
a
0

6
a
3

7
a
2

8
a
1
6
= 32
3
1
3
1
+24
3
1
1
+24p
1
2
1
2
1
+4p
1
2
1
a
1

7
a
3

8
a
2
7
= 24
2
1
2
1
+4
2
1
+8p
1
1
a
2

8
a
3
8
= p
1
+8
1
1
a
3
with p
1
,
1
,
1
> 0.
Now, consider that the system (6) is being perturbed by the unknown input signal (t) as
follows

1
=
2

2
=
3

3
=
4

4
= bu + (t)
y =
1
= L
(19)
with
(t) = a
0
1
a
1
2
a
2
3
a
3
4
+ (t)
Then, by the perfect knowledge of (t), one gets the following differential atness-based
controller with feedforward of the perturbation signal (t):
u =
1
b
( (t)) (20)
with
=
3
4

2
3

1
2

0
1
Similarly as before, the perturbation input signal could be locally reconstructed by the
following family of Taylor polynomial of (r 1)th degree:
(t) =
r1
i=0
i
t
i
(21)
where all the coefcients
i
are completely unknown.
Then, one can use the following extended mathematical model described in state-space
form to design a robust observer for real-time estimation of the disturbance (t) and time
derivatives of the at output required for implementation of the controller (20):

1
=
2

2
=
3

3
=
4

4
=
1
+ bu

1
=
2

2
=
3
.
.
.

r1
=
r

r
= 0
y =
1
= L
(22)
where
1
= ,
2
= ,
3
= , ,
r
=
(r1)
.
A Luenberger observer for the system (22) is then given by

1
=
2
+
r+3
(y y)

2
=
3
+
r+2
(y y)

3
=
4
+
r+1
(y y)

4
=
1
+ bu +
r
(y y)

1
=
2
+
r1
(y y)

2
=
3
+
r2
(y y)
.
.
.

r1
=
r
+
1
(y y)

r
=
0
(y y)
y =
1
(23)
The estimation error dynamics e
1
= y y satises the following dynamics:
e
(r+4)
1
+
r+3
e
(r+3)
1
+
r+2
e
(r+2)
1
+
r+1
e
(r+1)
1
+
r
e
(r)
1
+
r1
e
(r1)
1
+ +
2
e +
1
e
1
+
0
e
1
= 0
(24)
Therefore, the design parameters
i
, i = 0, , r + 3, can be chosen so that the output
estimation error e
1
exponentially asymptotically converges to zero.
On the other hand, it is assumed that the perturbation input signal (t) can be locally
approximated by a family of Taylor polynomials of fourth degree. Therefore, the characteristic
polynomial for the dynamics of output observation error (24) is given by
p
o2
= s
9
+
8
s
8
+
7
s
7
+
6
s
6
+
5
s
5
+
4
s
4
+
3
s
3
+
2
s
2
+
1
s +
0
(25)
It is then proposed the following Hurwitz polynomial to compute the proper gains for the
observer:
p
o2
(s) = (s + p
2
)
s
2
+2
2
2
s +
2
2
4
(26)
One then obtains that
0
= p
2
8
2
1
=
8
2
+8p
2
7
2
2
= 8
7
2
2
+24p
2
6
2
2
2
+4p
2
6
2
3
= 24
6
2
2
2
+4
6
2
+32p
2
5
2
3
2
+24p
2
5
2
4
= 32
5
2
3
2
+24
5
2
2
+16p
2
4
2
4
2
+48p
2
4
2
2
2
+6p
2
4
2
5
= 16
4
2
4
2
+48
4
2
2
2
+6
4
2
+32p
2
3
2
3
2
+24p
2
3
2
6
= 32
3
2
3
2
+24
3
2
2
+24p
2
2
2
2
2
+4p
2
2
2
7
= 24
2
2
2
2
+4
2
2
+8p
2
8
= p
2
+8
2
2
with p
2
,
2
,
2
> 0.
From the practical viewpoint, main advantage of this high-gain observer is that it could be
employed for hydraulic or electromagnetic active vehicle suspension systems, requiring only
information of the stiffness constant of tire k
t
and the unsprung mass m
u
. In addition, it
can be shown that the proposed observer design methodology is quite robust with respect to
parameter uncertainty and unmodeled dynamics, by considering the parameter variations
into the perturbation input signal (t). In fact, in (Sira-Ramirez et al., 2008a) has been
presented through some experimental results that the polynomial disturbance signal-based
GPI control scheme, implemented as a classical compensation network, is robust enough with
respect to parameter uncertainty and unmodeled dynamics in the context of an off-line and
pre-specied reference trajectory tracking tasks.
It is important to emphasize that, the proposedresults are nowpossible thanks to the existence
of commercial embedded systemfor automatic control tasks based on high speed FPGA/DSP
boards with high computational performance operating at high sampling rates. The proposed
observer could be implemented via embedded software applications without many problems.
5. Simulation results
Some numerical simulations were performed on a nonlinear quarter-vehicle suspension
system characterized by the following set of realistic parameters (Tahboub, 2005) to verify
the effectiveness of the proposed disturbance observer-control design methodology (see Table
1):
Parameters Values
Sprung mass (m
s
) 216.75 [kg]
Unsprung mass (m
u
) 28.85 [kg]
Spring stifness (k
s
) 21700 [
N
m
]
Damping constant (c
s
) 1200 [
Ns
m
]
Tire stifness (k
t
) 184000 [
N
m
]
nonlinear spring stiffness (k
ns
) 2170 [
N
m
]
nonlinear damping constant (c
ns
) 120 [
Ns
m
]
Table 1. Parameters of the vehicle suspension system.
Fig. 2 shows some schematic diagramfor the implementation of the proposed active vibration
controllers based on on-line disturbance estimation using a atness-based controller and GPI
observers.
Fig. 2. Schematic diagram of the instrumentation for active vehicle suspension control
implementation.
The following trajectory was utilized to simulate the unknown exogenous disturbance
excitations due to irregular road surfaces (Chen & Huang, 2005):
z
r
(t) =
f
1
(t) + f (t) for t [3.5, 5)
f
2
(t) + f (t) for t [5, 6.5)
f
3
(t) + f (t) for t [8.5, 10)
f
3
(t) + f (t) for t [10, 11.5
f (t) else
with
f
1
(t) = 0.0592 (t 3.5)
3
+0.1332 (t 3.5)
2
f
2
(t) = 0.0592 (t 6.5)
3
+0.1332 (t 6.5)
2
f
3
(t) = 0.0592 (t 8.5)
3
0.1332 (t 8.5)
2
f
3
(t) = 0.0592 (t 11.5)
3
0.1332 (t 11.5)
2
f (t) = 0.002 sin(2t) +0.002 sin(7.5t)
Figs. 3-9 describe the robust performance of the controller (7) using the observer (14). It can
be seen the high vibration attenuation level of the active vehicle suspension systemcompared
with the passive counterpart.
Moreover, one can observe a robust and fast on-line estimation of the disturbance (t) as well
as the corresponding time derivatives of the at output up to third order. Similar results on
the implementation of the controller (20) with disturbance observer (23) for estimation of the
perturbation (t) and time derivatives of the at output are shown in Figs. 10-23.
In the computer simulations it is assumed that the perturbation input signals (t) and (t)
can be locally approximated by a family of Taylor polynomials of fourth degree.
The characteristic polynomials for the ninth order observation error dynamics were all set to
be of the following form:
p
o
(s) = (s + p
o
)
s
2
+2
o
o
s +
2
o
4
with p
o
=
o
= 300rad/s and
o
= 20.
The characteristic polynomials associated with the closed-loop dynamics were all set to be of
the form: p
c
(s) =
s
2
+2
c
c
s +
2
c
2
, with
c
= 10rad/s and
c
= 0.7071.
0 5 10 15
0.15
0.1
0.05
0
0.05
0.1
0.15
t [s]
D
i
s
p
l
a
c
e
m
e
n
t

[
m
]

Active
Passive
Road profile
Fig. 3. Sprung mass displacement response using controller (7) and observer (14).
0 5 10 15
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
t [s]
A
c
c
e
l
e
r
a
t
i
o
n

[
m
/
s
2
]

Active
Passive
Fig. 4. Sprung mass acceleration response using controller (7) and observer (14).
In general, the proposed active vehicle suspension using a atness-based controller and GPI
observers for the estimation of unknown perturbations yields good attenuation properties and
an overall robust performance.
0 5 10 15
0.15
0.1
0.05
0
0.05
0.1
0.15
t [s]
D
e
f
l
e
c
t
i
o
n

[
m
]

Active
Passive
Fig. 5. Suspension deection response (x
1
x
3
) using controller (7) and observer (14).
0 5 10 15
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
x 10
3
t [s]
D
e
f
l
e
c
t
i
o
n

[
m
]

Active
Passive
Fig. 6. Tire deection response (x
3
z
r
0 5 10 15
2
1.5
1
0.5
0
0.5
1
1.5
2
x 10
7
t [s]
P
e
r
t
u
r
b
a
t
i
o
n

S
i
g
n
a
l

Estimated perturbation
Actual perturbation
Fig. 7. Perturbation estimation (t) using observer (14).
0 5 10 15
10
0
10
t [s]
F
i
r
s
t

d
e
r
i
v
a
t
i
v
e

0 5 10 15
200
100
0
100
200
t [s]
S
e
c
o
n
d

d
e
r
i
v
a
t
i
v
e

0 5 10 15
2
0
2
x 10
4
t [s]
T
h
i
r
d

d
e
r
i
v
a
t
i
v
e

Estimate Actual value
Fig. 8. Estimation of time derivatives of the at output using the observer (14).
0 5 10 15
2500
2000
1500
1000
500
0
500
1000
1500
2000
2500
t [s]
u

[
N
]
Fig. 9. Control force using the observer (14).
0 5 10 15
0.2
0.15
0.1
0.05
0
0.05
0.1
0.15
t [s]
D
i
s
p
l
a
c
e
m
e
n
t

[
m
]

Active
Passive
Road profile
Fig. 10. Sprung mass displacement response using controller (20) and observer (23).).
0 5 10 15
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
t [s]
A
c
c
e
l
e
r
a
t
i
o
n

[
m
/
s
2
]

Active
Passive
Fig. 11. Sprung mass acceleration response using controller (20) and observer (23).
0 5 10 15
0.15
0.1
0.05
0
0.05
0.1
0.15
t [s]
D
e
f
l
e
c
t
i
o
n

[
m
]

Active
Passive
Fig. 12. Suspension deection response (x
1
x
3
0 5 10 15
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
x 10
3
t [s]
D
e
f
l
e
c
t
i
o
n

[
m
]

Active
Passive
Fig. 13. Tire deection response (x
3
z
r
0 5 10 15
2
1.5
1
0.5
0
0.5
1
1.5
2
x 10
7
t [s]
P
e
r
t
u
r
b
a
t
i
o
n

s
i
g
n
a
l

Estimated perturbation
Actual perturbation
Fig. 14. Perturbation estimation (t) using observer (23).
0 5 10 15
10
5
0
5
10
t [s]
F
i
r
s
t

d
e
r
i
v
a
t
i
v
e

0 5 10 15
200
100
0
100
200
t [s]
S
e
c
o
n
d

d
e
r
i
v
a
t
i
v
e

0 5 10 15
1
0.5
0
0.5
1
x 10
4
t [s]
T
h
i
r
d

d
e
r
i
v
a
t
i
v
e

Fig. 15. Estimation of time derivatives of the at output using the observer (23).
0 5 10 15
2500
2000
1500
1000
500
0
500
1000
1500
2000
2500
t [s]
u

[
N
]
Fig. 16. Control force using the observer (23).
6. Conclusions
In this chapter a robust active vibration control scheme, based on real-time estimation
and rejection of perturbation signals, of nonlinear vehicle suspension systems is described.
The proposed approach exploits the structural property of differential atness exhibited
by the suspension system fot the synthesis of a atness based controller and a robust
observer. Therefore, a perturbed input-output differential equation describing the dynamics
of the at output is obtained for design purposes of the control scheme. The exogenous
disturbances due to irregular road surfaces, nonlinear effects, parameter variations and
unmodeled dynamics are lumped into an unknown bounded time-varying perturbation
input signal affecting the differentially at linear simplied dynamic mathematical model
of the suspension system. A family of Taylor polynomials of (r-1)th degree is used to
locally approximate this perturbation signal. Hence the perturbation signal is described by
a rth-order mathematical model. Then, the perturbed suspension system model is expressed
as a (r+4)th-order extended mathematical model.
The design of high-gain Luenberger observers, based on this kind of extended models, is
proposed to estimate the perturbation signal and some time derivatives of the at output
required for implementation of differential atness-based disturbance feedforward and
feedback controllers for attenuation of vibrations in electromagnetic and hydraulic active
vehicle suspension systems.
Two high-gain disturbance observer-based controllers have been proposed to attenuate the
vibrations induced by unknown exogenous disturbance excitations due to irregular road
surfaces, which could be employed for nonlinear quarter-vehicle active suspension models by
using hydraulic or electromagnetic actuators. Computer simulations were included to show
the effectiveness of the proposed controllers, as well as of the disturbance observers based on
Taylor polynomials of fourth degree.
The results show a high vibration attenuation level of the active vehicle suspension system
compared with the passive counterpart and, in addition, a robust and fast real-time estimation
of the disturbance and time derivatives of the at output.
7. References
Ahmadian, M. Active control of vehicle suspensions. In: Encyclopedia of Vibration, Edited by
Braun, S.G., Ewins, D.J. & Rao, S.S. (2001), Vols. 1-3, Academic Press, San Diego, CA.
Basterretxea, K., Del Campo, I. & Echanobe, J. (2010). A semi-active suspension embedded
controller in a FPGA, 2010 IEEE International Symposium on Industrial Embedded
Systems, pp. 69-78, Trento, July 7-9.
Beltran-Carbajal, F., Silva-Navarro, G., Blanco-Ortega, A. & Chavez-Conde, E. (2010a).
Active Vibration Control for a Nonlinear Mechanical Systemusing On-line Algebraic
Identication, In: Vibration Control, M. Lallart, (Ed.), 201-214, Sciyo, Rijeka, Croatia.
Beltran-Carbajal, F., Silva-Navarro, G., Sira-Ramirez, H. & Blanco-Ortega, A. (2010b).
Application of on-line algebraic identication in active vibration control, Computacin
y Sistemas, Vol. 13, No. 3, pp. 313-330.
Cao, J., Liu, H., Li, P. & Brown, D. (2008). State of the Art in Vehicle Active Suspension
Adaptive Control Systems Based on Intelligent Methodologies, IEEE Transaction on
Intelligent Transportation Systems, Vol. 9, No. 3, pp. 392-405.
Choi, S.B., Lee, H.K. & Chang, E.G. (2001). Field test results of a semi-active ER suspension
system associated with skyhook controller, Mechatronics, Vol. 11, pp. 345-353.
Chavez-Conde, E., Beltran-Carbajal, F., Garcia-Rodriguez, C. & Blanco-Ortega, A. (2009a).
Sliding Mode Based Differential Flatness Control and State Estimation of Vehicle
Active Suspensions System, IEEE International Conference on Electrical Engineering,
Computing Science and Automatic Control, pp. 544-549, Toluca, Mexico, November
10-13.
Chavez-Conde, E., Beltran-Carbajal, F., Blanco-Ortega, A. & Mendez-Azua, H. (2009b).
Sliding Mode and Generalized PI Control of Vehicle Active Suspensions, 18th
IEEE International Conference on Control Applications, pp. 1726-1731, Saint Petersburg,
Russia, July 8-10.
Chen, P. & Huang, A. (2005). Adaptive sliding control of non-autonomous active suspension
systems with time-varying loadings, Journal of Sound and Vibration, Vol. 282, pp.
1119-1135.
Fliess, M., Marquez, R., Delaleau, E. & Sira-Ramirez, H. (2002). Correcteurs
Proportionnels-Integraux Gnraliss, ESAIM Control, Optimisation and Calculus
of Variations, Vol. 7, pp. 23-41.
Fliess, M., Marquez, R. & Delaleau, E. (2001). State feedbacks without asymptotic observers
and generalized PID regulators, Nonlinear Control in the Year 2000, Lecture Notes in
Control and Information Sciences, Vol. 258, pp. 367-384, Springer, London.
Fliess, M., Lvine, J., Martin, Ph. & Rouchon, P. (1993). Flatness and defect of nonlinear
systems: introductory theory and examples, International Journal of Control, Vol. 61,
No. 6, pp. 1327-1361.
Gysen, B.L.J., Paulides, J.J.H., Janssen, J.L.G. & Lomonova, E. A. (2008). Active
Electromagnetic Suspension System for Improved Vehicle Dynamics, IEEE Vehicle
Power and Propulsion Conference (VPPC), Harbin, China, September 3-5.
Isermann, R. &Munchhpf, M. (2011). Identication of Dynamic Systems, Springer-Verlag, Berlin.
Martins, I., Esteves, J., Marques, D.G. & Da Silva, F. P. (2006). Permanent-Magnets Linear
Actuators Applicability in Automobile Active Suspensions, IEEE Trans. on Vehicular
Technology, Vol. 55, No. 1, pp. 86-94.
Sira-Ramirez, H., Beltran-Carbajal, F. &Blanco-Ortega, A. (2008a). AGeneralized Proportional
Integral Output Feedback Controller for the Robust Perturbation Rejection in a
Mechanical System, e-STA Sciences et Technologies de lAutomatique, Vol. 5, No. 4, pp.
24-32.
Sira-Ramirez, H., Feliu-Batlle, V., Beltran-Carbajal, F. & Blanco-Ortega, A. (2008b).
Sigma-Delta modulation sliding mode observers for linear systems subject to locally
unstable inputs, 16th Mediterranean Conference on Control and Automation, pp. 344-349,
Ajaccio, France, June 25-27.
Sira-Ramirez, H., Barrios-Cruz, E. & Marquez-Contreras, R.J. (2009). Fast adaptive trajectory
tracking control for a completely uncertain DC motor via output feedback,
Computacin y Sistemas, Vol. 12, No. 4, pp. 397-408.
Sira-Ramirez, H., Silva-Navarro, G. &Beltran-Carbajal, F. (2007). On the GPI balancing control
of an uncertain Jeffcot rotor model, 2007 4th International Conference on Electrical and
Electronics Engineering (ICEEE), pp. 306-309, Mexico City, Mexico, September 5-7.
Sira-Ramirez, H. & Agrawal, S.K. (2004). Differentially Flat Systems, Marcel Dekker, New York.
Shoukry, Y., El-Kharashi, M. W. & Hammad, S. (2010). MPC-On-Chip: An Embedded GPC
Coprocessor for Automotive Active Suspension Systems, IEEE Embedded Systems
Letters, Vol. 2, No. 2, pp. 31-34.
Tahboub, K. A. (2005). Active Nonlinear Vehicle-Suspension Variable-Gain Control, 13th
Mediterranean Conference on Control and Automation, pp. 569-574, Limassol, Cyprus,
June 27-29.
van der Schaft, A. (2000). L
2
Gain and Passivity Techniques in Nonlinear Control, Springer,
London.
Ventura, P.J.C., Ferreira, C.D.H., Neves, C. F. C. S., Morais, R.M.P., Valente, A.L.G. & Reis,
M.J.C.S. (2008). An embedded system to assess the automotive shock absorber
condition under vehicle operation, IEEE Sensor 2008 Conference, pp. 1210-1213, Lecce,
October 26-29.
Yao, G.Z., Yap, F.F., Chen, G., Li, W.H. & Yeo, S.H. (2002). MR damper and its application for
semi-active control of vehicle suspension system, Mechatronics, Vol. 12, pp. 963-973.
Kazuhiko Hiramoto
1
, Taichi Matsuoka
2
and Katsuaki Sunakoda
3
1
Department of Mechanical and Production Engineering, Niigata University
2
Department of Mechanical Engineering Informatics, Meiji University
3
Sanwa Tekki Corporation
*
Japan
1. Introduction
Various methodologies for vibration control of civil structures have been proposed so far.
The traditional scheme is passive vibration control, i.e., dissipation of the vibration energy to
the outside of the structural systems with dampers or mass dampers etc.. Passive control is
quite simple and popular still, however it has some limitations, e.g., insufcient performance
and/or difculty in tuning such devices for a case of multi-mode vibration control etc..
Active vibration control is a candidate for a breakthrough to overcome the above drawbacks
of passive control and has been studied extensively these decades ((Spencer et al., 1998) and
the references therein). Although many studies show that the active control methodology
achieves the quite high control performance on vibration suppression, it requires a large
energy source to produce the control force and this fact has been an obstacle in applying
active methods to general vibration control problems.
Semi-active control, which is not necessarily new (Karnop et al., 1974) either, can be
recognized as an intermediate between passive and active schemes in the sense of not only the
performance on vibration suppression but also the complexity of the control system. In most
semi-active control vibration suppression is achieved by changing the damping coefcient
of the semi-active damper. In civil structures semi-active control technique is getting more
realistic recently (Casciati et al., 2006) along with the development of a large scale damper
whose damping property is able to be changed (Sodeyama et al., 1997). In semi-active control,
the damping coefcient of the semi-active damper is changed mainly based on the following
two strategies:
An active controller is synthesized with a standard control theory (e.g., sky-hook, LQ or
H
control, etc.) rstly as a target for the semi-active control. Then the semi-active damper
is operated so as to emulate the targeted active control force as much as possible(Dyke et
al., 1996; Karnop et al., 1974; Kurata et al., 1999).
The damping coefcient of the semi-active damper is determined so as to maximally
accelerate energy dissipation of the structural systems(Gavin, 2001).
*
A professor emeritus at Akita University, Japan

Semi-Active Control of Civil Structures
Based on the Prediction of the Structural
Response: Integrated Design Approach

8
2 Will-be-set-by-IN-TECH
The latter method is based on the Lyapunov theory. The Lyapunov function has been dened
as the total energy, i.e., the sumof the kinetic and potential energies of the structural systemin
most cases. The control law for maximizing the rate of energy dissipation is generally given
as a bang-bang type change of the damping coefcient.
In this chapter we propose a newsemi-active control method that is based on a one-step-ahead
prediction of the structural response for the seismic disturbance. In the present semi-active
control the vibration control device (VCD), which has been developed by the authors to
reduce the structural vibration, is employed. The VCD is a mechanism consists of a ball
screw, a ywheel and an electric motor connected to the ball screw. The VCD is installed
between neighboring two oors of buildings as general dampers. Unlike general dampers,
the VCD generates two types of resistance forces, i.e., a damping force proportional to the
relative velocity and an inertial force proportional to the relative acceleration between two
oors. The damping force is generated by the electric motor and the inertial force is produced
by the mechanism of the ball screwand the ywheel. The damping coefcient of the VCDcan
be adjusted in a real-time manner by changing the electric resistance connected to the motor
and we use this capability of the VCD for the semi-active control.
We assume the command signal for changing the damping coefcient of each VCD takes
two values, i.e., the command to take the maximum or minimum damping coefcient in the
present chapter. Under the assumption we obtain the optimal command signal among all
possible combinations of command signals (2
n
VCD
, n
VCD
: The number of VCDs) in a real-time
manner. The optimal command signal is selected from all candidates of the command signals
so that the weighted Euclidean norm of the predicted one-step-ahead structural response,
calculated by a numerical integration method, e.g., Runge-Kutta method, is minimized.
To accomplish the further performance improvement of the semi-active control system an
integrated design approach of structural and semi-active control systems using the above
predictive control law is proposed. The stiffness distribution between neighboring two oors
of the structure and some parameters of the VCDs are taken as adjustable structural design
parameters. The weighting matrix that is used in the predictive semi-active control law
is dened as the control design parameters. Those design parameters are simultaneously
optimized so that the seismic responses of the semi-active control system subject to various
recorded and articial earthquake waves become good. The control performance is evaluated
by an objective function that is calculated by the simulated structural response of the
semi-active control system. The Genetic Algorithm (GA) is adopted for the optimization.
As a design example the a simulation study for a fteen-story building with three VCDs
is presented. The control performance of the semi-active control system is signicantly
improved by the proposed integrated design methodology.
The rest of the chapter is organized as follows: In 2 the principle of the VCD and the
mathematical model of the civil structure with the VCD are introduced. The semi-active
control law based on the prediction of the structural response is given in 3. The integrated
design problem of the structural and the semi-active control is formulated in 4. The
simulation results for a fteen-story building are shown in 5 and the conclusion of the chapter
is presented in 6.
Notations are as follows: t: time, R
mn
: the set of m n real matrices, R
m
: the set of
m-dimensional real vectors, S
n
: the set of n-dimensional real symmetric matrices, 0
mn
: an
mn zero matrix, I: an identity matrix with an appropriate dimension, 1
mn
: an mn matrix
whose all elements equal to 1,
T
: the transposition of a vector or a matrix.
Semi-active Control of Civil Structures Based on the Prediction of the Structural Response: Integrated Design Approach 3
2. Mathematical model
2.1 Vibration Control Device (VCD)
The schematic diagram of the VCD is shown in Fig. 1. The VCD is installed between
neighboring two oors of a structure, e.g., the i-th and (i 1)-th oors. In Fig. 1, q
i
(t), m
VCD
j
and d
VCD
j
(t) are the displacement of the i-th oor, an equivalent mass and a time-varying
damping coefcient of the VCD
j
(j-th VCD, j = 1 . . . , n
VCD
) respectively. The resistance
force produced by the relative motion of the two oors is denoted by f
VCD
j
(t). The VCD
is a dynamic system which has two input signals, i.e., the relative acceleration and velocity
between two oors. The output signal of the VCD is the resistance force f
VCD
j
(t). The
mathematical model of the VCD is described as the following:
f
VCD
j
(t) = m
VCD
j
( q
i
(t) q
i1
(t)) + d
VCD
j
(t)( q
i
(t) q
i1
(t)) (1)
The VCD whose resistance force property is given as Eq. (1) has been developed by the
authors in recent years (Ohtake et al., 2006). The assembly and the design parameters of
the VCD in (Ohtake et al., 2006) are shown in Fig. 2 and Table 1 respectively. The VCD is
a mechanism consists of a ball screw, a ywheel, an electric motor and an electric circuit to
control the damping property of the VCD. The ball screw and the ywheel produce a large
inertia force (the term m
VCD
j
( q
i
(t) q
i1
(t)) in Eq. (1)) which is proportional to the relative
acceleration between two oors. The damping force denoted by d
VCD
j
(t)( q
i
(t) q
i1
(t)) in
Eq. (1) is generated by the electric motor (generator) and the electric circuit for the energy
dissipation. The coefcients m
VCD
j
and d
VCD
j
(t) in Eq. (1) are given as (Ohtake et al., 2006):
m
VCD
j
= K
s
(I
1
+ r
2
I
2
), d
VCD
j
(t) = K
s
r
2
K
t
K
e
R
a
+ R
, (2)
where K
s
, r, I
1
and I
2
are the constant related to the ball screw, the gear ratio, the moment of
inertia of the ywheel and that of the electric motor respectively. In the damping coefcient
d
VCD
j
(t), K
t
, K
e
, R
a
and R are the torque constant, the back-emf (back electromotive force)
constant, the electric resistance of the motor and the resistance connected to the motor terminal
respectively. From Eq. (2) the damping coefcient of the VCD can be controlled by changing
the resistance R. We have developed an electronic circuit to change the electronic resistance
according to a command voltage that is produced by an implemented semi-active control law.
More detailed dynamical properties of the VCD are presented in (Ohtake et al., 2006) and we
are currently developing the VCD that can produce a larger resistance force.
2.2 Model of civil structures with VCD
Consider a civil structure with the n
VCD
VCDs, whose inertial and variable damping
coefcients are denoted by m
VCD
j
and d
VCD
j
(t) (j = 1, . . . , n
VCD
) respectively. The equation of
motion of the structure is given as
M q(t) +D(t) q(t) +Kq(t) = b
2
w(t) +b
1
(t) w(t) +b
0
w(t), (3)
M = M
0
+M
VCD
, D(t) = D
0
+D
VCD
(t), b
1
(t) = b
0
1
+b
VCD
1
(t), b
2
= b
0
2
+b
VCD
2
, (4)
where q(t) R
n
and w(t) R
n
w
are the displacement and the disturbance vectors of
the structure respectively. Matrices M, D, K S
n
and b
j
R
nn
w
(j = 0, 1, 2) are the
153
Based on the Prediction of the Structural Response: Integrated Design Approach
Fig. 1. Schematic diagram of the VCD
Design parameters Value [Unit]
Length 0.699 [m]
Weight 16.8 [kg]
Maximum load 30 [kN]
Stroke 6.0 10
2
[m]
Table 1. Design parameters of the VCD(Ohtake et al., 2006)
mass, damping, stiffness and inuence coefcient matrices of the structure with the VCDs
respectively. In Eq. (4) the matrices M
0
, D
0
, b
0
1
and b
0
2
are the mass, damping and inuence
coefcient matrices of the original structure without VCDs and the matrices M
VCD
, D
VCD
(t),
b
VCD
1
(t) and b
VCD
2
are those of the installed VCDs respectively. Note that the matrices D and
b
1
accordingly become time varying because they contain the variable damping coefcient of
the VCD.
With the equation of motion in Eq. (3) the state-space form of the structure with n
VCD
VCDs
is given by
_
x(t) = Ax(t) +Bu
x
(t)
z(t) = C
z
x(t) +D
z
u
z
(t)
, (5)
where
x(t) :=
_
q(t)
q(t)
_
, u
x
(t) :=
w(t)
w(t)
w(t)
, u
z
(t) :=
_
w(t)
w(t)
_
,
A :=
_
0
nn
I
n
M
1
K M
1
D(t)
_
, B :=
_
0
n3n
w
M
1
_
b
0
b
1
(t) b
2
_
.
The vector z(t) R
n
z
is the output vector to evaluate the performance of the structural system
with VCDs. The output vector z(t) will be used to construct the semi-active control law that
will be proposed in the next section. Note that all coefcient matrices in Eq. (5) can become
time varying because possibly they contain the variable damping coefcient of the VCDs,
i.e., matrices A, B, C
z
, D
z
are functions on the variable damping coefcient (d
VCD
j
(t), j =
1, . . . , n
VCD
) of the VCDs. In this chapter we assume each variable damping coefcient of the
VCD can be controlled in a following range:
d
VCD
j
d
VCD
j
(t) d
VCD
j
, j = 1, . . . , n
VCD
(6)
where d
VCD
j
0 and d
VCD
j
0 are the maximum and minimum damping coefcients of the
j-th VCD respectively.
Fig. 2. Assembly of the VCD(Ohtake et al., 2006)
3. Semi-active control methodology
We propose a semi-active control using the variable damping capability of the VCD. The
control law is based on a prediction of the seismic response of the structure. The prediction
of the structural response can be obtained through a numerical integration of the state-space
model dened in Eq. (5) with the current sensor data of the state variable and the earthquake
disturbance. The prediction of the structural response is obtained for all candidates of
command signals each VCD. Then the optimal (combination of) command signal is selected
among all candidates of the command signals in a real time manner. Essentially the command
signal for the VCD can take a continuous value in the range given as Eq. (6). In this chapter,
however, to make the prediction-based semi-active control realistic, the command signal for
each VCD denoted by d
c
j
(t), j = 1, . . . , n
VCD
is quantized to the following one bit signal:
d
c
j
(t) =
d
VCD
j
d
VCD
j
, j = 1, . . . , n
VCD
(7)
With the above quantization the possible number of command combinations for n
VCD
VCDs
becomes 2
n
VCD
. In the present chapter the optimal combination is selected among all possible
combinations so that the following performance index J
p
is minimized:
J
p
:=
_
z
T
(t
0
+t)Q
w
z(t
0
+t) (8)
where t
0
and t > 0 are the current time instant and the length of the prediction respectively.
The matrix Q
w
= Q
T
w
0 is an weighting matrix, dened by a control system designer, for
evaluating the predicted control performance.
From Eq. (5) the prediction of the output vector z(t
0
) is given as
z (t
0
+t) = C
z
x (t
0
+t) +D
z
u
z
(t
0
+t) , (9)
where the prediction of the state vector x(t
0
+ t) and the output vector z(t
0
+ t) are
approximated by
x (t
0
+t) x (t
0
) +t x (t
0
) , u
z
(t
0
+t) u
z
(t
0
) +t u
z
(t
0
) . (10)
To have the (approximated) predicted output vector z (t
0
+t) we need to obtain x (t
0
) and
u
z
(t
0
). The time derivative of the state vector can be obtained with the current state vector
x (t
0
) and the disturbance u
x
(t
0
) at t = t
0
by using the state-space form Eq. (5), that is,
x (t
0
) = Ax (t
0
) +Bu
x
(t
0
). (11)
155
We can apply various numerical integration methods, e.g., Euler method or Runge-Kutta
method etc., to obtain (the approximated) x
0
(t
0
). Furthermore the time derivative of the
disturbance u
z
(t
0
) can be obtained from the following simple relationship:
u
z
(t
0
) =
d
dt
_
w(t
0
)
w(t
0
)
_
=
_
w(t
0
)
w(t
0
)
_
(12)
In the present chapter all 2
n
VCD
values of J
p
s, the weighted Euclidean norm of the predicted
output vector z(t
0
+t), are computed for all possible combinations of the command signals
for the VCDs at every time instant. Then the optimal combination that minimizes J
p
in Eq. (8)
is selected among all 2
n
VCD
candidates of command signals.
Similar to currently available devices that realize a variable damping property, such as MR
damper etc., the VCD in the present chapter also has a dynamic delay for a given command
signal. In other words the i-th damping coefcient of the VCD, denoted by d
VCD
i
(t), cannot
perfectly track the command signal d
c
i
(t) dened as Eq. (7). We assume the characteristic of
the delay is modeled as the following rst order dynamics:
d(d
VCD
j
(t))
dt
=
1
T
d
d
VCD
j
(t) +
1
T
d
d
c
j
(t), j = 1, . . . , n
VCD
(13)
where T
d
> 0 is the time constant representing the delay of the VCD.
Several features of the proposed prediction-based semi-active control are summarized as
follows:
In the semi-active control the predicted performance index J
p
in Eq. (8) is obtained for
all possible combinations of the command signals and the optimal command that achieves
the minimum J
p
is selected. As shown in 1 conventional semi-active control laws consider
the norm of the closed-loop system (in the phase of the targeted active control design) or
the energy dissipation rate of the structural systems. Such performance indices are used
in general control system design (not only in semi-active control design) as a measure
for evaluating the closed-loop performance. However in the semi-active control of civil
structures we can not know if a satisfactory result is obtained or not with the adopted
semi-active control in the sense of the vibration control of civil structures when we employ
one of the conventional semi-active control methods because a mismatch exists between
the above performance index and key performance measures used in vibration control of
civil structures, e.g., the relative displacement between neighboring two oors and the
absolute acceleration of each oor. In other words there is a gap between the performance
index for the control system design and the quantities for the performance evaluation in a
semi-active control of civil structural systems. On the other hand the performance index
J
p
is the Euclidean norm of the predicted output vector z(t) in Eq. (5). The elements of the
vector z(t) can be dened as the relative displacement and velocity between neighboring
two oors and the absolute displacement and velocity of each oor, etc. (Note: Clearly,
there is a strong relationship between the absolute acceleration and the displacement or
velocity of each oor.). Therefore the gap mentioned above is small in the proposed control
method. By comparing the proposed prediction-based approach with the conventional
semi-active control laws in 1 the prediction-based method is a more direct method
because the performance measure on vibration suppression of the civil structural systems
can be directly dealt with to obtain the optimal command for the VCDs in a real-time
manner.
The proposed semi-active control is quite exible because it is a fully algorithm-based
control law. As the most noteworthy example of the exibility we show that a robust
performance property within the present semi-active control framework can be recovered
easily with a simple method that is shown as follows. In general model-based semi-active
control methods including the proposed prediction-based approach the performance of
the control system depends on the accuracy of the model of the control object. Because
the perfect mathematical model of the control object is generally unavailable we need
to make the control law to be robust in some sense. In the present semi-active control
the predicted performance index J
p
in Eq. (8) is obtained for all 2
n
VCD
candidates of the
command signals and the optimal command that achieves the minimum J
p
is selected. Let
J
max
p
denote the predicted performance index obtained by setting the command signals
for all VCDs to their maximum values, i.e., d
c
i
(t) = d
VCD
i
, i = 1, . . . , n
VCD
. As a simple
method to recover the robustness we redene the performance index J
max
p
as J
max
p
,
0 < 1 and then the optimal command signal is selected in the same way. After
the redenition the command signal corresponding to (the redened) J
max
p
tends to be
selected more often because the command d
c
i
(t) = d
VCD
i
, i = 1, . . . , n
VCD
is overvalued
among all candidates of the optimal command signals. By making 0 < 1 smaller, the
performance of the semi-active control approaches asymptotically to that the case where all
damping coefcients are kept their maximum. Actually in = 0 the semi-active control is
coincident with the passive control where all damping coefcients are xed their maximum
because the command signal d
c
i
(t) = d
VCD
i
, i = 1, . . . , n
VCD
is always selected during
the semi-active control as a result. The above methodology is reasonable because the
maximum damping case guarantees a certain level of vibration suppression independent
of the mathematical model of the structural system although the resulted performance of
the semi-active control becomes conservative. Note that in general control design there is
a trade-off between the performance and the robustness like this situation. As the authors
knowledge, among so many studies about semi-active control of civil structures classied
in 1 there are no theoretical considerations about the robustness of the semi-active control
system although all model-based semi-active control laws highly rely on the mathematical
model of the structural systems. The authors would like to emphasize the fact that
we should not expect the semi-active control methodologies aiming to realize the desired
active control as much as possible, e.g., clipped optimal control(Dyke et al., 1996), have a
performance robustness comparable with the targeted active control even if the targeted
active control law is robustly designed in some sense. It is because the semi-active devices
can not produce the same control force as the targeted active control perfectly. With
the above discussion we can say that the present approach is simple but useful because
the performance robustness of the present semi-active control can be easily recovered by
making the factor smaller and the degree of the robustness also can be adjusted by tuning
the parameter .
As other examples of the exibility of the control law we present following two features:
A constraint on the hardware can be included quite easily, e.g., a constraint on the
command signal such that all commands for VCDs must take the same value at some
instant.
In the present study we use the Euclidean norm(2-norm) of the predictedoutput. Other
norms such as 1-norm and -norm also can be used.
157
Additionally, with the bang-bang control nature of the method the implementation of the
control law is relatively easier than general feedback control methods that require the
continuous controller output.
4. Integrated design approach
Under the predictive semi-active control law in 3, an integrated design problemof structural
and control systems is formulated. The integrated design problemis the control systemdesign
problem that the controller and the structural parameters, e.g., the stiffness and the damping
characteristics of structural systems, are dealt with as adjustable design parameters.
The integrated design problem was rstly considered in the control system design of large
space structures (LSS) in 1980s and has been studied around this three decades in the control
system and the structural design communities(Grogoriadis et al., 1996; Hiramoto et al., 2000;
Hiramoto & Grigoriadis, 2006; Onoda & Haftka, 1987). In the integrated design paradigm we
can expect better control performance than that of the standard structural or control design
because we can adjust the design parameters simultaneously, i.e., we can use the full design
freedom of the formulated design problem. However, we currently have the result that the
general integrated design problem becomes a BMI (Bilinear Matrix Inequality) optimization
problem that cannot be obtained the globally optimal solution with the practically acceptable
amount of computation even in the simplest case. Several methods have been proposed so far
to obtain a locally optimal solution(Grogoriadis et al., 1996; Hiramoto et al., 2000; Hiramoto &
Grigoriadis, 2006; Onoda & Haftka, 1987).
In this chapter the integrated design problemof the civil structural systemwith the predictive
semi-active control presented in the previous section is formulated. As the structural design
parameters we take the structural stiffness between neighboring two oors denoted by k
i
, i =
1, . . . , n, the equivalent mass and the maximum value of the variable damping coefcient of
each VCD that are denoted by m
VCD
j
and d
VCD
j
, j = 1, . . . , n
VCD
respectively. Each structural
design parameter is able to be adjusted in the ranges as follows:
(k
i
)
l
k
i
(k
i
)
u
, i = 1, . . . , n (14)
_
m
VCD
j
_
l
m
VCD
j

_
m
VCD
j
_
u
, j = 1, . . . n
VCD
(15)
_
d
VCD
j
_
l
d
VCD
j

_
d
VCD
j
_
u
, j = 1, . . . n
VCD
(16)
Note that subscripts
u
and
l
mean that the maximum and minimum values of the structural
design parameters respectively.
The control design parameters are dened as [Q
w
]
kk
, k = 1, . . . , n
z
that are the diagonal
elements of the weighting matrix Q
w
in Eq. (8). The range of [Q
w
]
kk
, k = 1, . . . , n
z
is given by
0 ([Q
w
]
kk
)
l
[Q
w
]
kk
([Q
w
]
kk
)
u
, k = 1, . . . , n
z
. (17)
The objective function in the present integrated design problem is dened as the following:
J = J
s
+ (1 ) J
h
, 0 < < 1, (18)
J
s
=
n
e
i=1
J
s
i
, J
s
i
=
4
j=1
g
j
i
(J
s
i
)
j
, g
j
i
0, i = 1, . . . , n
e
, j = 1, . . . , 4, (19)
(J
s
i
)
1
=
n
k=1
RMS
_
s
r
i
k
(t)
_
RMS
_
0
r
i
k
(t)
_
, (20)
(J
s
i
)
2
=
n
k=1
RMS
_
s
a
i
k
(t)
_
RMS
_
0
a
i
k
(t)
_
, (21)
(J
s
i
)
3
=
n
k=1
max
0tt
f
s
r
i
k
(t)
max
0tt
f
0
r
i
k
(t)
, (22)
(J
s
i
)
4
=
n
k=1
max
0tt
f
s
a
i
k
(t)
max
0tt
f
0
a
i
k
(t)
, (23)
J
h
=
n
e
i=1
J
h
i
, J
h
i
=
4
j=1
h
j
i
_
J
h
i
_
j
, h
j
i
0, (24)
(J
h
i
)
1
=
n
k=1
RMS
_
h
r
i
k
(t)
_
RMS
_
0
r
i
k
(t)
_
, (25)
(J
h
i
)
2
=
n
k=1
RMS
_
h
a
i
k
(t)
_
RMS
_
0
a
i
k
(t)
_
, (26)
(J
h
i
)
3
=
n
k=1
max
0tt
f
h
r
i
k
(t)
max
0tt
f
0
r
i
k
(t)
, (27)
(J
h
i
)
4
=
n
k=1
max
0tt
f
h
a
i
k
(t)
max
0tt
f
0
a
i
k
(t)
, (28)
where n
e
and T
i
f
, i = 1, . . . , n
e
are the number and the duration time of the i-th earthquake
waves that are used to obtain the simulated structural response in the present integrated
design respectively. The relative displacement between neighboring k-th and (k 1)-th oors
and the absolute acceleration of k-th oor are denoted by r
i
k
and a
i
k
, i = 1, . . . , n
e
, k = 1, . . . , n
respectively. Superscripts
s
,
h
and
0
show the case of semi-active control, the case where the
damping coefcient of all VCDs are kept their maximum throughout the earthquake event
(Passive on) and the case without VCDs. The scalar 0 < < 1 and g
j
i
, i = 1, . . . , n
e
are
weighting factors that are determined by the designer.
The components J
s
i
, i = 1, . . . , n
e
are the sum of ratios between the RMS or the peak values of
structural responses (r
i
k
and a
i
k
, i = 1, . . . , n
e
, k = 1, . . . , n) that are obtained respectively in the
semi-active control case and the passive on case for the i-th earthquake wave. Hence, if we
get a solution to the integrated design problem that the component J
s
i
, i = 1, . . . , n
e
is small,
the amount of the performance improvement is large when the predictive semi-active control
law is applied.
159
On the other hand the components J
h
i
, i = 1, . . . , n
e
are the sum of ratios between the RMS
or the peak values of r
i
k
and a
i
k
, i = 1, . . . , n
e
, k = 1, . . . , n in the passive on case and the case
without VCDs. When the component J
h
i
, i = 1, . . . , n
e
is small, the performance enhancement
by installing VCDs (and keeping the damping coefcients maximum) is large.
The performance index J in Eq. (18) is the weighted sum of J
s
representing the performance
improvement with the semi-active control and J
h
related to the effect of the VCD itself.
By minimizing the performance index J we can obtain the structural and control design
parameters that achieve the good control performance on vibration suppression.
The formulated integrated design problem that minimizes the objective function J in Eq. (18)
is a highly nonlinear and non-convex optimization problem on structural and the control
design parameters because of the form of the objective function itself and the nonlinear
nature of the semi-active control law. Furthermore the number of the design parameters is
relatively high in the formulated optimization problem compared to the standard structural
or the control system design problem because we need to optimize not only the structural
or control design parameters only but parameters both in the structural and control system
simultaneously. Therefore it is difcult to apply gradient-based optimization methods that
achieve a local convergence of the objective function J because the amount of computation to
get the gradient of J on the structural and control design parameters become enormous. In
the present chapter the Genetic Algorithm (GA) is applied to obtain an optimal solution. In
GA the global optimality of the obtained solution is not guaranteed. However no gradient
information is required to get the solution and many successful applications of GA have been
reported so far for various types of optimization problems.
5. Design example
Let us consider a fteen-story building with three VCDs shown in Fig. 3. The one-dimensional
earthquake disturbance whose displacement is dened as w(t) is assumed in the present
example. The mass of each oor m
i
and the nominal inter-story stiffness k
nom
i
, i = 1, . . . , 15 are
shown in Table 2. In the present design example the structural damping matrix D
0
is dened
as
D
0
= K, > 0, (29)
where the scaler > 0 is determined so that the damping ratio of the 1st mode of vibration
becomes 2%. The nominal parameter values of the VCDs are determined as
_
m
VCD
j
_
nom
= 9.0 10
6
[kg], (30)
_
d
VCD
j
_
nom
= 3.6 10
7
[Ns/m],
_
d
VCD
j
_
nom
= 0, j = 1, 2, 3. (31)
The maximum and minimum values of the structural design parameters in Eqs. (14)-(16) are
dened as follows:
(k
i
)
l
= 0.5k
nom
i
, (k
i
)
u
= 2k
nom
i
, i = 1, . . . , 15 (32)
_
m
VCD
j
_
l
= 0.5
_
m
VCD
j
_
nom
,
_
m
VCD
j
_
u
= 2
_
m
VCD
j
_
nom
, (33)
_
d
VCD
j
_
l
= 0.5
_
d
VCD
j
_
nom
,
_
d
VCD
j
_
u
= 2
_
d
VCD
j
_
nom
, j = 1, 2, 3 (34)
Fig. 3. Fifteen story building with 3-VCDs
Story m
i
10
3
[kg] k
nom
i
10
6
[N/m]
15 882 291
14 470 294
13 470 305
12 470 313
11 470 352
10 480 412
9 485 427
8 485 435
7 485 448
6 485 483
5 500 544
4 505 561
3 505 580
2 505 644
1 900 1098
Table 2. Nominal structural parameters of the fteen story building (i = 1, . . . , 15)
The control output vector z(t) in Eq. (5) is dened as
z(t) =
r(t)
r(t)
q(t)
q(t)
= C
z
x (t) +D
z
u
z
(t) , (35)
161
Design parameter [Unit] Nominal Optimal
m
VCD
1
[kg] 9.00 10
6
5.04 10
6
m
VCD
2
[kg] 9.00 10
6
4.78 10
6
m
VCD
3
[kg] 9.00 10
6
4.78 10
6
d
VCD
1
[Ns/m] 3.60 10
7
6.79 10
7
d
VCD
2
[Ns/m] 3.60 10
7
6.78 10
7
d
VCD
3
[Ns/m] 3.60 10
7
7.04 10
7
Table 3. Nominal and optimal design parameters of VCDs
where q(t) and r(t) are the vector of the absolute displacement of each oor and the relative
displacement between neighboring two oors given as
q(t) =
q
1
(t)
q
2
(t)
.
.
.
q
15
(t)
, r(t) =
r
1
(t)
r
2
(t)
.
.
.
r
15
(t)
q
1
(t) w(t)
q
2
(t) q
1
(t)
.
.
.
q
15
(t) q
14
(t)
. (36)
Then coefcient matrices C
z
and D
z
in Eq. (5) are given by
C
z
=
R 0
1515
0
1515
R
I
30
, D
z
=
s 0
151
0
151
s
0
3030
(37)
where
R =
1 0 0 0 . . . 0
0 1 1 0 . . . 0
0 0 1 1 . . . 0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0 . . . . . . . . . 1 1
, s =
_
1
0
141
_
. (38)
The time constant T
d
in Eq. (13) to model the delay of the VCD for the command signal d
j
c
(t),
j = 1, 2, 3 is dened as T
d
= 0.02 [s].
From the denition of z(t) in Eq. (35) the weighting matrix Q
w
in Eq. (8) has 60 diagonal
elements. Because the number of elements is too many to conduct the GA-based optimization
the matrix Q
w
is parameterized as
Q
w
= diag
_
w
l
r
1
15
, w
m
r
1
15
, w
h
r
1
15
, w
l
dr
1
15
, w
m
dr
1
15
, w
h
dr
1
15
,
w
l
q
1
15
, w
m
q
1
15
, w
h
q
1
15
, w
l
dq
1
15
, w
m
dq
1
15
, w
h
dq
1
15
_
, (39)
where w
l
r
, w
m
r
, w
h
r
, w
l
dr
, w
m
dr
, w
h
dr
, w
l
q
, w
m
q
, w
h
q
, w
l
dq
, w
m
dq
, w
h
dq
> 0 are the weighting factors
for the relative displacement and velocity for lower (1st-5th), middle (6th-10th) and higher
(11th-15th) oors, and those for the absolute displacement and velocity for the oors divided
like the above way respectively. With the parameterization the number of the control design
parameters becomes 12. The maximum and minimum values of those weighting factors are
determined to be 1 and 0 respectively.
To obtain the simulated structural responses that are used in the proposed GA-based optimal
design we take four recorded or articial earthquake waves (n
e
= 4 in Eqs. (19) and (24)), i.e.,
El Centro NS (1940) (i = 1), BCJL1 (articial) (i = 2), Hachinohe NS (1968) (i = 3) and JMA
kobe NS (1995) (i = 4) waves. All earthquake waves are scaled such that the peak ground
acceleration (PGA) becomes 4.0 [m/s
2
].
The optimal structural and control design parameters are obtained by minimizing the
objective function J in Eq. (18) with GA. The optimized design parameters are shown in Figs.
4, 5 and Table 3. To compare the performance on vibration suppression between the optimal
semi-active control system obtained by the present integrated design approach and some
other cases, structural responses are obtained for the scaled BCJL2 (articial, PGA: 4.0 [m/s
2
])
earthquake wave
1
shown in Fig. 6 for seven cases given as follows:
NC (Nominal structure): The case of the nominal structural design parameters without
VCD
NC (Optimized structure): The case of the optimized structural design parameters
without VCD
Pon (Nominal structure): The passive on case with the nominal structural design
parameters
Pon (Optimized structure): The passive on case with the optimized structural design
parameters
SA (Nominal structure): The semi-active control case with the nominal structural and
control design parameters
SA (Optimized structure): The semi-active control case with the optimized structural and
control design parameters
Energy-based SA (Optimized structure): The case of the semi-active control for
the maximum energy dissipation(Gavin, 2001) with the optimized structural design
parameters
The time histories of the structural responses r
15
(t) (the relative displacement between the
15th and 14th oors) and a
15
(t) (the absolute acceleration of the 15th oor) for the optimized
structure are presented in Fig. 7. By comparing between the case NC, Pon and SA the
structural responses are suppressed by introducing the VCDs. It is also clear that the further
performance improvement is achieved with the proposed predictive semi-active control by
comparing the responses of the case Pon and SA. The time histories of the variable damping
coefcients of the three VCDs are also shown in Fig. 8. The damping coefcients of the
VCDs are frequently changed between their maximumand minimumvalues by the predictive
semi-active control. The intermediate values between the maximum and minimum values of
the variable damping coefcients come from the delay of the VCDs modeled as Eq. (13).
To make a more quantitative evaluation of the control performance the RMS and the
peak values of the relative displacement between neighboring two oors and the absolute
acceleration of each oor for all the above seven cases are shown in Figs. 9-12. The results of
the integrated design for the fteen story building are summarized as follows:
1. By introducing VCDs the structural response for the earthquake disturbance is highly
suppressed both in the nominal and the optimized structures. This result shows that the
effectiveness of the VCD as the vibration control device for civil structures.
1
Note that this earthquake wave is not employed in the GA-based optimization.
163
0 2 4 6 8 10 12
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Initial structure
Optimized structure
|N/m| 10
8
F
l
o
o
r
Fig. 4. Nominal and optimal stiffness parameters k

k
, k = 1, . . . , 15
2. In the case Pon and the case SA the relative displacement r
k
, k = 1, . . . , 15 and the
absolute acceleration a
k
, k = 1, . . . , 15 of the optimized structure are signicantly improved
compared to those of nominal structure especially in middle and higher oors.
3. In lower oors (1-3F) the values of the optimized stiffness k
k
becomes much lower
than those of the nominal values while the stiffness in upper oors are increased. The
result implies that so called soft-story type structure (Iwata et al., 1999), that aims at
concentrating the structural deformation in lower oors by adopting small stiffness k
k
and moreover suppressing the vibration and the deformation in higher oors by adopting
large stiffness k
k
, is obtained as the result of the integrated design. Hence the relative
displacement in lower oors becomes larger in the case NC for the optimized structure
than that of the nominal structure. However, we can see that the large amount of the
relative displacement in the lower oors of the optimized structure are well suppressed by
the VCDs with the proposed predictive semi-active control.
4. From Figs. 9-12 the optimized semi-active control system (SA (Opt. str.)) shows
the better performance on vibration suppression compared to that of the semi-active
control proposed by (Gavin, 2001) (Energy-based SA (Opt. str.))
2
. Especially the fairly
better control performance on the absolute acceleration of each oor is achieved with the
predictive semi-active control compared to that achieved with the he semi-active control
law proposed in (Gavin, 2001). This result indicates that the importance of the proposed
integrated design approach, i.e., the design of the structural system (the control object)
with taking the employed control law into consideration.
2
In the semi-active control law in (Gavin, 2001) the command signal for the VCD also becomes the
bang-bang type that is same as the present predictive semi-active control.
Fig. 5. Optimal control design parameters
Fig. 6. BCJL2 wave (PGA=4.0 [m/s
2
])
165
Fig. 7. Time histories of r
15
and a
15
for BCJL2 wave (Optimized structure, PGA=4.0 [m/s
2
])
Fig. 8. Time histories of the variable damping coefcients of VCDs for BCJL2 wave
(Optimized structure PGA=4.0 [m/s
2
])
Fig. 9. RMS values of the relative displacement RMS(r
k
), k = 1, . . . , 15 (BCJL2 wave)
167
Fig. 10. RMS values of the absolute acceleration RMS(a
k
) , k = 1, . . . , 15 (BCJL2 wave)
Fig. 11. Peak values of the relative displacement max |r
k
|, k = 1, . . . , 15 (BCJL2 wave)
169
Fig. 12. Peak values of the absolute acceleration max |a
k
|, k = 1, . . . , 15 (BCJL2 wave)
6. Conclusion
In this chapter the integrated design of civil structural systems and the semi-active control law
is considered. The vibration control device (VCD) that is under development by the authors
is adopted for the semi-active control. The VCD is the mechanism consists of the ball-screw
and the ywheel that is for the inertial resistance force and the electric motor and the electric
circuit for the damping resistance force with the variable damping coefcient. The semi-active
control based on the one-step ahead prediction of the structural response is proposed. With
the predictive semi-active control the stiffness property of the building, the parameters of
VCDs and the weighting matrix used in the semi-active control are simultaneously optimized
so that the control performance on vibration suppression for various recorded and articial
earthquake disturbances is optimized. The Genetic Algorithmis adopted for the optimization.
The simulation study is conducted for the fteen story building. The performance on vibration
suppression of the semi-active control system obtained by the integrated design method is
veried with the earthquake wave that is not employed in the GA-based optimization. The
result of the simulation study shows the effectiveness of the proposed design methodology
and the importance of the integrated design approach for control system design including
semi-active control.
The future research subjects are summarized as follows:
Integrated design of the semi-active control system including the optimization of the
location of the VCDs
Semi-active control for a simplied model of the real structural system
Experimental study using a full-scale building with VCDs
7. References
Casciati, F., Magonette, G., & Marazzi, F. (2006). Technology of Semiactive Devices and Application
in Vibration Mitigation, Wiley, New York.
Dyke, S.J., Spencer, Jr., B.F., Sain M.K. & Carlson J.D. (1996). Modeling and control of
magnetorheological dampers for seismic response reduction, Smart Materials and
Structures, Vol. 5: 565-575.
Gavin H.P. (2001). Control of seismically excited vibration using electrorheological materials
and Lyapunov methods, IEEE Transactions on Control Systems Technology, Vol. 9: 27-36.
Grigoriadis, K.M., Zhu, G. & Skelton, R.E. (1996). Optimal Redesign of Linear Systems,
Transactions of the ASME, Journal of Dynamics, Systems, Measurement and Control, Vol.
118: 598-605.
Hiramoto, K., Doki, H. & Obinata, G. (2000). Optimal Sensor/Actuator Placement for Active
Vibration Control Using Explicit Solutions of Algebraic Riccati Equation, Journal of
Sound and Vibration, Vol. 299: 1057-1075.
Hiramoto, K. & Grigoriadis, K.M. (2006). Integrated Design of Structural and Control Systems
with a Homotopy Like Iterative Method, International Journal of Control, Vol. 79:
1062-1073.
Karnopp, D., Crosby, M.J., & Harwood, R.A. (1974). Vibration control using semi-active force
generator, Transactions of the ASME, Journal of Engineering for Industry, Vol. 96: 619-626.
Kurata, N., Kobori, T., Takahashi M., Niwa N. & Midorikawa H. (1999). Actual seismic
response controlledbuilding with semi-active damper system, Earthquake Engineering
& Structural Dynamics, Vol. 28: 1427-1447.
171
Ohtake T., Sunakoda K. & Matsuoka, T. (2006). Study on vibration control device using power
generator, Proceedings of ASME PVP 2006, Vancouver, #PVP2006-ICPVT-11-93534.
Iwata, N., Hama, T. & Soda, S. (1999). Seismic control of the soft-rst-story building by sifting
method in sliding mode control, Journal of Structural and Construction Engineering,
(No. 816): 83-90.
Onoda, J. &Hattka R.T. (1987). An Approach to Structure/Control Simultaneous Optimization
for Large Flexible Spacecraft, AIAA Journal, Vol. 25: 1133-1138.
Sodeyama, H., Suzuki, K. & Sunakoda, K. (2004). Development of large capacity semi-active
seismic damper using magneto-rheological uid, Transactions of the ASME, Journal of
Pressure Vessel Technology, Vol. 126: 105-109.
Spencer, B.F., Dyke, S.J. & Deoskar, H.S. (1998). Benchmark problems in structural control:
Part I - Active mass driver system, Earthquake Engineering & Structural Dynamics, Vol.
27: 1127-1139.
9
Seismic Response Control Using
Smart Materials
Sreekala R, Muthumani K, Nagesh R Iyer
CSIR/Structural Engineering Research Centre
India
1. Introduction
Earthquakes are highly destructive natural phenomena resulting in the massive
deterioration of civil infrastructure, which becomes highly significant wih increasing urban
population. Performance of structural systems needs to be improved recalling the huge loss
of life and destruction to constructed facilities caused by various earthquakes. Protection of
lifelines and infrastructural facilities are of utmost importance during a seismic event. There
have been considerable research efforts in seismic vibration control for the past several
decades. Developments of new techniques and new materials, which are not traditionally
used in civil engineering structures, offer significant promise in reducing the seismic risk.
Smart materials may be described as materials that can sense an external stimulus (e.g:-
stress, pressure, temperature change, magnetic field, etc.) and initiate a response. They may
belong to one of the four classes namely metals or alloys, polymers, ceramics or composites.
Metals and alloys of different metals are considered as classical materials with lot of
research activities around the globe . Shape Memory Alloys (SMA) belongs to the class of
smart materials which are well known for its peculiar characteristics, which can be stress or
temperature, induced. The research area which deals with structural applications of this
variety of smart materials are promising (Sreekala& Muthumani, 2009) for structural health
monitoring and vibration control.
Nickel Titanium (NiTi) Alloys are well known for its super elastic and shape memory
properties and they belong to the class of Shape Memory Alloys (SMA). Presently SMAs are
mainly applied in medical sciences, electrical, aerospace and mechanical engineering and
also can open new applications in civil engineering specifically in seismic protection of
buildings. Super elastic nitinol is found to be very effective for passive vibration control, as
it can sustain large amounts of inelastic deformation and recover that deformation at the
end of the process with good energy dissipation compared to regular metallic materials.
Various tests are conducted on NiTi wires in CSIR-SERC to evaluate the possibility of using
SMA as an energy dissipating material with re-centering capabilities. Ability to sustain and
recover large amounts of inelastic deformation during reversed cyclic loading with extra
ordinary fatigue resistance make super elastic NiTi suitable candidate for seismic risk
reduction. Quasi static and dynamic tests conducted with various parameters like pre strain,
amplitudes and frequency for a number of cycles establishes the behavior which is suitable
for seismic applications. It is interesting to find that from the quasi-static behavior of the
material an optimum value of pre strain, which is material dependent, can be selected and


174
adopted in vibration control applications. Mathematical models were developed to predict
the maximum energy dissipation capability of the material under study. The tests verify the
potential of pre-strained NiTi wires as kernel components in seismic protection devices. It
also highlights the effects of repeated cycling on pre strained wires, as special devices for
passive protection during earthquake are meant for repeated cycling. Various types of
vibration control devices can be made with the material used.
The technologies using smart materials are useful for both new and existing constructions.
The chapter highlights various structural applications of this class of smart materials
available in addition to the suitability of this material for vibration control applications. .
Protection of structures from damage during earthquakes can be addressed using passive
protection devices designed to have maximum energy dissipation capacity with re centering
capabilities. Innovative means of structural control scheme can be developed through
passive systems, consisting of materials and devices useful in the context of seismic design
and retrofit of structures
Ductility and energy dissipation are important parameters to be evaluated and quantified
for seismic resistant design of structural components. Equivalent viscous damping is a
significant quantity usually employed in earthquake engineering to characterize the
dissipation capability of the material under investigation. This parameter expresses the
effectiveness of the material in vibration damping. The relevance of the study is made clear
with the associated terms in earthquake engineering
2. Properties, which enable SMA for civil engineering application
1. Repeated absorption of large amounts of strain energy under loading without
permanent deformation
2. Possibility to obtain a wide range of cyclic behaviour from supplemental and fully re
centering to highly dissipating by simply varying the characteristics of SMA components.
3. Strain range of 2 to 10%
4. Extraordinary fatigue resistance under large strain cycles.
5. Their greater durability and reliability in the long run.
2.1 Shape memory alloys in structural control
The existence of SMA property in certain alloys has been known from 1932(when it was first
observed in Gold-Cadmium alloy exhibits a rubber like behavior). SMAs have unique
properties, which are not present in many materials traditionally used in engineering
applications; especially shape memory effect and super elasticity. While shape memory
alloys have been commercially available since the 1960s, their application has been limited
(Pelton et al., 2000). Early literatures on SMA materials shows successful uses of these for
the design of biomedical and clinical devices, particularly for fixing body fractures
internally with various NiTi SMA interlocking intramedullary nails(Da et al,2001). Among
other designs, research related to the use of SMA materials as micro-pumps, micro-grippers,
flexible SMA active catheters, micro nano-tribological devices and SMA thin films for
electromechanical system have been increasingly revealed in recent years (Stockel & Melzer,
1995). Those achievements exhibited an excellent reliability and confidence in using these
materials for other primarily structural elements. The SMA materials possess many unique
mechanical and geometrical properties such as shape memory effects which provide a large
recovery strain and force simultaneously, pseudo-elastic behavior upon heating and super-

Seismic Response Control Using Smart Materials

175
elastic properties in ambient conditions for specified SMA materials. Birman (1997) and
Humbeeck (2001) gave comprehensive reviews on the properties of SMA materials. Based
on the studies under MANSIDE project, 1999(Italian Department for National technical
services) NiTi SMAs appear to be the best candidates for use in seismic applications. But
selection of the SMA wires for energy dissipation with or without pre strain had to be
addressed separately (Sreekala et al., 2010). So the need arose for finding the effect of pre
strain as well as repeated cycling in the energy dissipation of SMA wires.
2.2 Mechanism behind super elasticity
Super elasticity is related to the isothermal response of SMA specimens to applied
mechanical loads. The phase transformations are solid to solid diffusionless processes
between a crystallographically more ordered phase, the austenite (A) and a
crystallographically less ordered phase the martensite (M). Fig. 1 shows a schematic stress
strain cyclic curve of a super elastic SMA. It is characterized by five branches. Branches 1
and 4 correspond to the elastic deformation of the two stable phases of SMA, respectively,
austenite and martensite. Branches 2 and 3 correspond, respectively, to the forward (from
austenite to detwinned martensite) and inverse (from detwinned martensite to austenite)
phase transformation. Branch 5 corresponds to the onset of plastic deformation of
detwinned martensite. In the figure o
Fs
and o
Ff
represent the critical stresses at which the
forward transformation, respectively, starts and finishes, while o
Is
and o
If
are those at which
the inverse transformation, respectively, starts and finishes. Since loading and unloading
paths are different from each other, a certain amount of energy is dissipated over the cycle.
3. Research significance
Most of the applications of SMA have focused on developing various types of actuators.
However, most of the commercially successful applications - such as couplings, electrical
connectors, cardiovascular stents and eyeglass frames - use the full shape- memory effect or
the full super elastic cycle only a few times. This indicates that SMA components can be
used successfully for several non-actuator applications as well. One such area is the use of
SMA components for damage control in structures. The SMA component can undergo finite
deformations either due to super-elasticity or due to rearrangement of martensite variants at
relatively small stresses. This can be used to prevent permanent plastic deformation due to
unexpected loads in the rest of the structure. The ability of the SMA component to recover
its original dimensions may be an additional advantage, during the seismic events.
Shape Memory Alloys (SMA) show the potential to eliminate some limitations involved in
present technologies, allowing a broader application. The limitations of existing
technologies for passive protection devices can be summarized as follows
1. Problems related to ageing and durability (e.g., rubber components)
2. Difficulty in maintenance(those based on fluid viscosity)
3. Installation complexity or replacement and geometry restoration after strong events
(those based on steel yielding or lead extrusion)
4. Variable performances depending on temperature(polymer based devices)
The usable strain range in the order of 10% in shape memory alloys provide them very high
energy dissipation per unit mass of material. Comparing the well known rubber isolators
and steel hysteretic dampers which belong to the class of quasi-elastic devices and elasto-
plastic devices respectively, re-centering devices gain the best mechanical characteristics of


176
both. The shortcomings of the traditional restrainers can potentially be addressed with the
use of SMAs (Fig.2). The comparative behaviour of steel, rubber and SMA is given in Fig.2.

Fig. 1. Stress-strain curve showing super elasticity of SMA-Schematic sketch describing the
phase change with the deformation process and the associated crystal structure
On the one hand, they recover the initial position of the structure, with a good control of the
displacements; on the other hand they put a threshold to the force transmitted to the super-
structure. The full possibility of designing the mechanical behavior combining the self-
centering and the energy dissipation capability, permit to calibrate the desired features and
fit the specific needs. The availability of such features opens considerable room for
improvement of the structural system design. But the mechanical behavior of SMAs is
strongly dependent upon the alloy composition and the thermo-mechanical treatment.
Special care should be taken while alloy selection is made.


177
SMA
STEEL
RUBBER

Fig. 2. Comparative behaviour of Steel, Rubber and SMA
4. Experimental investigation and inferences
The experimental tests were carried out on austenite wire samples of 0.4mm diameter and
1.2mm. The wires were available in spool form. The typical photograph of the material
wires in spool form-is shown in fig.3. The compositions of wires were NiTi alloy with 55 %
Ni and balance titanium spooled NiTi. The wire selection is made in such a way that Ni-Ti
alloy wire with equi-atomic composition possesses better dissipation property and higher
resistance to corrosion and fatigue. The material is straight annealed super elastic NiTi wires
and has its latent heat and specific gravity 14500 J/kg and 6479.85 kg/m
3
(0.234 lbs/in
3
)
respectively. The test was carried out under ambient temperature conditions around 27
0
C.
Quasi-Static and dynamic tests were conducted to evaluate the super-elastic properties and
energy dissipation capabilities under ambient temperature. Dynamic tests were carried out
by applying sinusoidal cyclic deformation to wire samples. A displacement controlled test
set up was used to perform the tests. The typical test sequence is characterized in terms of
number of cycles, strain amplitude and frequency as they could provide very important
information regarding earthquake applications.
The test programme on SMA wires of two diameters namely 0.4 and 1.2 mm can be
summarized as follows:
1. Quasi-static tests
2. Dynamic tests -frequency of operation 0.5 Hz (sinusoidal cyclic deformation)
a. Tests to evaluate the number of cycles to failure for a constant amplitude
b. Variable amplitude tests to evaluate the energy dissipation capability for increased
amplitudes.
3. Tests on pre-strained wires frequency of operation 0.5 Hz .
4. Tests on wires for varying frequencies.
The quasi-static (slow rate) tests were performed till the failure of the specimen. The test
was run in a position controlled test set up at constant rate of loading 0.025mm/sec. Fig.4
shows the stress strain curve for quasi-static loading. It is compared with the behavior of
the 1mm diameter steel wire which is used for binding purposes in construction A
sinusoidal cyclic test on SMA wires was performed for maximum amplitude of 5 mm.


178

Fig. 3. The photograph of the super elastic SMA wire samples (Available in spool form)

-0.02 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16
0
200
400
600
800
1000
1200
NiTi-SMA-0.4mm Dia.
Mild Steel-1mm Dia.
S
t
r
e
s
s
-
M
P
a
Strain

Fig. 4. Behaviour of Ni-Ti wires quasi static tests- Comparison with 1 mm mild- steel
binding wire used for reinforcement stirrup tying in structural construction.
The mechanical behavior such as the transformation stress levels, fatigue behaviour and the
frequency dependence were found to be similar in both diameter wires. The tests were also
carried out with increasing number of cycles to obtain the fatigue characteristics.
Typical load deformation characteristics of 1.2mm diameter wire with 11 mm amplitude are
shown in Fig.6. As the displacement amplitude increased as observed in the force
deformation behaviour, presence of the martensitic phase -after completion of the phase
transformation- provides additional stiffness. Strain hardening is observed towards higher
amplitudes which correspond to 8-9% strain in both the wires seen in Fig.4 & Fig.6. The
increase in stress levels is quite evident in Fig.6. Here cycling up to 9% strain variation
fetches a stress of 830 Mpa.


179

Fig. 5. Load deformation curves for 0.4mm diameter wire with 5mm amplitude (strain
variation 3%, maximum stress 650 Mpa)- frequency of loading 0.5 Hz-0.4mm diameter
wires

0
20
40
60
80
100
-5 0 5 10 15
Deformation(mm)
L
o
a
d
(
K
g
)
11 mm cycles

Fig. 6. Load-deformation curves for 1.2 mm diameter wire with 11mm amplitude cycles
(strain range 9%, max. stress 830MPa)- frequency of loading 0.5 Hz Cycles (strain range 9%,
max. stress 830MPa)
The various parameters and the testing scheme are selected in a way to understand the
behavior of the material during earthquakes. Initially a sinusoidal cyclic test on SMA wires
was performed for constant amplitude of 5 mm [ Fig.7(a)- 7(d)]. Usually earthquake


180
vibrations are dynamic- reversed cyclic in nature which consists many number of cycles.
The tests were also carried out with increasing number of cycles to obtain the fatigue
characteristics. Further tests were carried out with increasing amplitude from 5 mm to 11
mm [i.e increasing the total strains from 4% to 9%, Fig.7 (e)]. Strain measurements were
based on the actuator displacement as the wires were having very smooth oxide finish
surface

(a)

(b) Initial 28 cycles (c) Subsequent cycles


181

(d) (e)
Fig. 7. (a) to (d) Cyclic tests on non pre-strained wires cycling for +/- 3 percentage strain
variation (5mm amplitude). (e) Effect of increasing the amplitude beyond 1500 cycles.
Due to the repeated cyclic deformation of the same amplitude it is observed that the
hysteresis loops translate downwards and narrows (Fig.7a-7d). Consequently the stress
levels for the transformation decreases. Energy dissipation capacity under constant
amplitude loading remained same up to 500 cycles, 20% reduction was observed in the next
500 cycles and around 30% reduction towards the failure from the initial value. Effect of
increasing the amplitude beyond 1500 cycles can be observed in figure.7e.
4.1. Effect of pre-strain
The effect of pre-strain was found out by imposing pre-strain amplitude in the wire. The
wire is held at that point of pre-strain and reversed-cycling performed for various
amplitudes (Fig.8b). Like wise for various pre strain values the dynamic tests were
performed by giving sinusoidal deformation to the wire samples (Fig.8a). These tests help to
simulate the characteristics of earthquake loading and understand the behaviour of the
material. The frequency dependence on loading is also investigated during the experiment
(Fig.9). The frequency of loading was varied from 0.5 Hz to 3 Hz, at intervals of 0.5 Hz, since
most of the earthquake applications come within this range. Table.1 gives the test
programme for the nitinol wire and the parameters evaluated in the study are tabulated.
Temperature variation during cycling was observed using a thermocouple for quite a few
cycles.
5. Discussion of test results
Quasi-static tests of the wire clearly shows the super elastic property of the SMA wire. Fig.4.
presents the stress strain curves obtained under quasi-static loading using a hydraulically
driven test system (instron make). The shape of the stress strain curve exhibits some of the
fundamental characteristics of SMA. Test result shows the super elastic property of the SMA
wires in the strain range 2% to 10 %. The initial elastic to plateau transition, that is the


182
transition from linearly elastic to super-elastic occurs at a stress of approximately 597 MPa.
The yield stress of 597 Mpa and ultimate stress of 1100 Mpa is comparable with the
idealized behavior of SMA (Dolce, 1994).

(a)
0 2 4 6 8 10 12 14
0
2
4
6
8
10
12
14
16
Pre-Strain- 7%
1mm@0.5Hz
2mm@0.5Hz
3mm@0.5Hz
5mm@0.5Hz
7mm@0.5Hz
L
o
a
d

(
k
g
)
Displacement (mm)

(b)
Fig. 8. Typical sketch showing the loading and the hysteretic curves obtained at 7% pre-
strain variable amplitudes.


183
If the material is unloaded after being loaded into the super elastic region, the unloading
portion of the stress strain curve does not follow the loading portion, but follows a lower
path back to the origin which indicates the transformation of stress induced martensite back
to the austenite (Fig.5,6). This makes the hysteretic energy dissipation which is an important
parameter to be assessed for earthquake applications and hence in passive control devices.
The super elasticity based applications take advantage of the following features 1) the
possibility of recovering large deformations 2) the existence of a transformation stress
plateau, which guarantees constant stress over non-negligible strain intervals. The
frequency dependence on loading is also investigated during the experiment (Fig.9). The
frequency of loading was varied from 0.5 Hz to 3 Hz, at intervals of 0.5 Hz, since most of the
earthquake applications come within this range. It is interesting to note that the number of
cycles for the same amplitude is not sensitive to the frequencies in the range 0.5-3 Hz as
observed in both the diameter wires.

Fig. 9. Typical Cyclic behavior (tension compression tests) of non pre-strained wires for
varying frequencies.
It is reported in the literature that the effect of frequency is negligible in the energy
dissipation in the range of interest 0.5 - 3 Hz for seismic applications (Cardone et al,1999) .
But it is observed from the test that the dissipated energy during the sinusoidal loading at
0.5 Hz is 25 % more than the energy dissipated at 1Hz to 3Hz, which is almost a constant
value. Effect of increasing the amplitude beyond 1500 cycles can be observed in figure.7e.
The high fatigue resistance combined with negligible residual deformation, observed in
SMA material help to reduce the post earthquake repair costs, if dampers are made out
using these.
It is found from the experiment that the hysteresis loops narrow and translate upwards
when there is an increase in strain amplitude, while the branches of the curve relevant to the


184
phase transformations harden, thus yielding an increase in the stress levels (Sreekala et
al,2008). This trend is observed in the case of cycling around 7% pre-strain, as the strain
reaches to 10% the end of the transformation range (Fig.8). This phenomenon is related to
the elastic deformation of the de-twinned martensite found at the end of the phase
transformation. This is true for pre strained wires as well as without pre strained. It is a
favorable aspect in seismic applications as the system stiffens rather than softening, if the
expected design seismic action is exceeded, thus ensuring a good control of displacements.
Constant amplitude tests for many cycles for varying frequencies indicated that there is
slight variation in the shape of the loops. Different heat exchange modalities with ambient
temperature, at different strain rates are the causes of these effects from quasi-static to
dynamic range (0.5-3 Hz). As strain rate increases, the heat-exchange condition increasingly
digresses from the isothermal one. The latent heat of transformation causes the specimen to
self-heat and then increase its average temperature during test. At the same time, during
each loading cycle, the specimen temperature oscillates around an average value, according
to the sinusoidal variation of strain. The temperature variation is found to be 9
0
c.While the
instantaneous temperature of the wire oscillates around the ambient temperature (27
0
c) with
excursions of the order of 9
0
c, the average temperature remains practically unchanged
during all the loading history equal to the ambient temperature.
As a result of the self heating (cooling) there occurs instantaneous increase (decrease) of the
stress required for transformation. It is observed from the test that the dissipated energy
during the sinusoidal loading at 0.5 Hz is 25 % more than the energy dissipated at 1Hz to
3Hz, which is almost a constant value. The hardening of the transformation branches, the
narrowing of the cycle, apparent change in the shape of the loop resulting in a reduced
energy loss all of which are associated with slight temperature variation due to different
heat exchange modalities are caused by the stabilization of the material due to repeated
cyclic loading(Fig.10). Possible increase of average temperature during cycling at higher
strain rates produces an upward translation [Miyazaki et al,1986b].
Stabilization of the material behavior due to repeated cyclic loading also can be observed
from the shape change of the loops in the tests on non-pre strained wires (Fig.10). The effect
of pre-strain is found at 2.5% pre strain applied in the form of deformation to the sample, on
two cyclic strains namely 4 % and 6 % (Fig.11.).
The figure reveals the fact that the hysteresis loops translate upwards as it near the
transformation stress level. The comparison between the energy dissipation capability at
various pre strain under different amplitudes, namely 3mm and 7mm is shown in Fig.12.
It is interesting to find that the quasi-static curve serves as the envelope curve in this process
and as and when the stress reaches the transformation stress level, as marked in fig.12,
further stiffening happens in the material. This is an added advantage for seismic
applications that the material gets stiffen, rather softening to have good control of
displacements. Fig. 12 also shows the hysteresis curve for the wire pre strained at 6% and
cycled between 2 % and 10% along with other strain variations.
Tests on wires pre strained to different levels, indicated a particular value of pre-strain,
from which cycling in the pseudo-elastic range fetches maximum energy dissipation (Fig.8.).
The equivalent viscous damping expresses the effectiveness of the material in vibration
damping. It is calculated based on the average energy dissipated per cycle and the damping
values observed are also found to be significant. Looking into 6% pre strain one can see the
higher equivalent damping values. Cyclic variation of 4% strain on either direction at 6%
pre strain gives higher equivalent damping due to higher energy dissipation (Table.1). An


185

0 2 4 6
0
2
4
6
8
10
1.5Hz-1
st
Cycle
L
o
a
d
(
k
g
)
Deformation(mm)
0 2 4 6
0
2
4
6
8
10
1.5 Hz.-500
th
Cycle
L
o
a
d
(
k
g
)
Deformation(mm)

(a) (b)

0 2 4 6
0
2
4
6
8
2Hz-1450
-th
cycle
L
o
a
d
(
k
g
)
Deformation(mm)
(c)
Fig. 10. (a) to (c) Changes in the shape of the curve during varying frequencies.


186

Fig. 11. Effect of pre-strain on the energy dissipation (1.2 mm diameter wires)- frequency of
loading 0.5 Hz

-0.02 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14
0
200
400
600
800
1000
1200
Quasi-Static
7%-3mm
7%-7mm
6%-7mm
5%-3mm
4%-3mm
3%-2mm
2%-3mm
2%-2mm
S
t
r
e
s
s
(
M
P
a
)
Strain

Fig. 12. Cyclic behavior of pre strained wires for varying pre strains. [The legend denotes
pre strain %-reverse displacement cycles from that prestrain (mm)]


187
average equivalent damping ratio of 16% was realized during the test [ Sreekala et al, 2008].
It is found to be nearer to half the super-elastic range obtained from quasi-static tests.
Results indicate that if the super elastic wires have to play an energy-dissipating role,
cycling around the threshold pre-strain value within this range is desirable [ Sreekala et
al,2010]. In this case the energy dissipation is dependent on the strain value in the super-
elastic range from where cycling the entire range fetches maximum, as seen in figure 4.
For wires without pre strain, the equivalent damping ratio observed is in the order of 3%.
This is because during cycling, the reverse direction (compressive region) does not
contribute for the energy dissipation in non pre-strained wires. Tests on virgin wires shows
lesser damping and energy dissipation capacity and these results can be compared with
those under MANSIDE project. The design of the SMA based devices can be made by
choosing the mechanical properties such as transformation stress levels/ plateau strength as
design variables. Fig.14 shows the hysteretic behaviour of the different types of devices that
can be developed using the properties of the SMA material. A variety of hysteretic
behaviour can be obtained from the material tested and its application can be made suitable
for seismic devices like re-centering, supplemental re-centering or in the case of non re-
centering devices. These results reveal the suitability of using pre strained SMA wires in
vibration control applications, especially for Non Re-Centering Devices-NRCD as seen in
Fig.14. Normally pre straining of wires in vibration control devices requires highly skilled
labour and maintenance becomes a difficult procedure. As far as the numbers of cycles are
concerned, pre strained wires also goes beyond 900 cycles under constant amplitude
loading. e.g.:-in 6% pre strained wire, under the maximum pseudo-elastic range goes up to
940 cycles before failure. No reduction in the energy dissipation is observed up to 30 cycles,
but towards the failure 30 % reduction is observed as in non-pre strained wire. Usually for
seismic applications, the numbers of cycles to be considered are less than 500 cycles as per
various codal spectrums for seismic design in different countries. Table 1 summarizes the
number of cycles, energy dissipation capacity per cycle, the strain variation and the
equivalent damping. These results reveal the suitability of using SMA wires in seismic
vibration control applications.
The usable strain range of the order of 10% in the alloys provide them very high energy
dissipation per unit mass of material. Re-centering devices gain the best mechanical
characteristics of both quasi-elastic devices (e.g., rubber isolators) and elasto-plastic devices
(e.g., steel hysteretic dampers). On the one hand, they recover the initial position of the
structure, with a good control of the displacements; on the other hand they put a threshold
to the force transmitted to the super-structure. The full possibility of designing the
mechanical behavior combining the self-centering and the energy dissipation capability,
permit to calibrate the desired features and fit the specific needs. The availability of such
features opens considerable room for improvement of the structural system design. Fig 7
indicates the various device options for passive vibration control using this material. But the
mechanical behavior of SMAs is strongly dependent upon the alloy composition and the
thermo-mechanical treatment. Special care should be taken while alloy selection is made.
6. Equivalent viscous damping
Some significant quantities usually employed in earthquake engineering to characterize the
dissipation capability of the material under investigation are equivalent viscous damping,
secant stiffness and energy loss per cycle. The equivalent viscous damping expresses the


188
effectiveness of the material in vibration damping. It is calculated based on the average
energy dissipated per cycle. It is calculated as

eq.
= W
D
/(2H K
S
o
2
) (1)

-0.02 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14
0
200
400
600
800
1000
1200
Stress in which elastic deformation of Martensite starts
684 MPa
Quasi-static
5%-3mm
6%-7mm
7%-3mm
7%-7mm
S
t
r
e
s
s
,
M
P
a
Strain

Fig. 13. Cyclic behavior of pre strained wires maximum energy dissipation when pre-
strained almost at the middle of the pseudo-elastic range. [The legend denotes pre strain %-
displacement cycles (mm)]

(1) (2) (3)

Fig. 14. Load deformation behavior (Hysterisis curves) for various types of Devices based
on SMA, SRCD-Supplementary Re-Centering Device, NRCD-Non Re-centering Devices, and
RCD-Re-centering Devices


189
where W
D
is the energy loss per cycle and o is the maximum cyclic displacement under
consideration. The secant stiffness (K
s
) is computed as
K
s
= (F
max
- F
min
) / (o
max
- o
min
) where F
max
and F
min
are the forces attained for the
maximum cyclic displacements o
max
and o
min
. The associated energy dissipation and the
equivalent viscous damping calculated are included in Table.1.
The effect of pre-strain is found at 2.5% pre strain applied in the form of deformation to the
sample, on two cyclic strains namely 4 % and 6 % (Fig.11.). An increase in the energy
dissipation can be observed as the cyclic strain amplitude increases. The energy dissipated
during cycling in pre-strained wires is considerably high compared to non-pre strained
wires as observed from the experiment. It is interested to observe that (Fig.13) leaving the
flat plateau; the response follows the elastic curve below 2% strain as obtained from the
quasi-static test. Hence the quasi- static behaviour gives the envelope curve for the cyclic
actions and a variety of hysteretic behaviour suitable for seismic devices can be obtained
with pre straining effect [Sreekala et al, 2010].
7. Modeling the maximum energy dissipation for the material under study
It is observed from the experiment that the wires pre strained to the middle of the strain
range gives the maximum energy dissipation(fig. 8). The behaviour can be predicted using
the following mathematical expressions.
{1/ }
n
E
Y
o |
c o
o c -
(
| |
= + (
|
\ .
(

(2)
{1/ }
n
E Y E
o o | o
c
c
-
(
| |
= + (
|
\ .
(

(3)

*
denotes the stress in the pre strained wire at the beginning of the cycling. Here can be
expressed as
{ ( ) }
.
in c
T
E f erf u | o c c oc c c
( | |
= +
( |
\ .
(4)
denotes the one dimensional back stress, Y is the yield stress ,that is the beginning of the
stress- induced transition from austenite to martensite, n is the overstress power. In
equation (4) the unit step function activates the added term only during unloading
processes. In the descending curve it contributes to the back stress in a way that allows for
SMA stress-strain description. The term with (.) shows the ordinary time derivative.
( )
y
y
E
E E
o =
is a constant controlling the slope of , where E is the elastic modulus of

austenite and E
y
is the slope after yielding.
Inelastic strain,
in
c ,is given by

in
E
o
c c = (5)


190
The error function, erf(x) ,and the unit step function, u(x) are defined as follows.

2
2
( )
t
erf x e dt
=
[
}
(6)
( ) 1 u x = , x 0 (7)
u(x) = 0 , x < 0 (8)
The basic expression obtained here for pre-strained wires is a modified form of the model
suggested by wilde et al, [12] which is used for predicting the tensile behaviour of SMA
materials. This mathematical expression describes the mechanical response of materials
showing hysteresis.

. . .
n
E
Y
o |
c
o c
(
| |
= (
|
\ .
(

(9)
where is the one dimensional stress and is the one-dimensional strain, is the one
dimensional back stress , E elastic modulus , Y the yield stress and n the constant controlling
the sharpness of the transition from elastic to plastic states.
The modified Cozzarelli model suggested by Wilde et al represents the hardening of the
material after the transition from austenite to martensite is completed. Here in this case for
pre strained wires for maximum energy dissipation, the hardening branch has not been
considered which requires additional terms containing further unit step functions. The
model is rate and temperature independent. The requirement of zero residual strain at the
end of the loading process motivated the selection of the particular form of back stress
through the unit step function. The coefficients f
t
, and c are material constants controlling
the recovery of the elastic strain during unloading. Since the stress and strain were found to
be independent of the rate, the time differential can be eliminated. Hence equation (2) and
(3) can be used for predicting the behaviour after many cycles by appropriately changing
the values of n and the starting stress value which follows the trend as in fig. 15. The
material constants for the wires tested were obtained as
f
t
= 0.115, C =0.001, n = 0.25, = 0.055. Using Equations (2) and (3) the curves are
fitted. Fig. 15 gives the fitted curve for the maximum energy dissipation for the material
tested.
8. Various seismic response mitigation strategies
Earthquake engineering has witnessed significant development during the course of the last
two decades. Seismic isolation and energy dissipation are proved to be the most efficient
tools in the hands of design engineer in seismic areas to limit both relative displacements as
well as transmitted forces between adjacent structural elements to desired values. Parallel
development of new design strategies (the seismic software) and the perfection of suitable
mechanical devices to implement the strategies (the seismic hardware) made it possible to
achieve efficient seismic response control. An optimal combination of isolator and the
energy dissipater ensures complete protection of the structure during earthquake. Energy
dissipation and re-centering capability are the two important functions to cater this need.


191
Most of the devices now in practice have poor re-centering capabilities. Instead of using a
single device a combination of devices can provide significant advantages
The chart shown below Fig.18 illustrates the various seismic response mitigation strategies.

0
200
400
600
800
1000
1200
0 0.02 0.04 0.06 0.08 0.1 0.12
strain
s
t
r
e
s
s
(
N
/
m
m
2
)
Experimental
Curve'"
Fitted Curve

Fig. 15. Cyclic behaviour of prestrained wires maximum energy dissipation obtained and
the fitted curve.

Fig. 16. Various seismic response mitigation strategies


192
Sl.no:
Nature of
Test
Pre- strain in
the wire
Cycling
Strain range
(%)
Average energy
dissipated per
cycle X10
-2
J
No: of
cycles
Equivalent
viscous
damping
1
Dynamic
Tests

Freq. of
operation
0.5 Hz.
(Sinusoidal
cyclic)
7%
6.5-7.5
6.0-8.0
5.0-9.0
4.0-10.0
3.0-11.0
1.3-12.0
00.30
00.60
06.70
18.60
33.00
-
10
10
a

10
a

10
a

10
a

-
0.02
0.02
0.06
0.07
0.09
-
2 6% 2-10
33.80
27.00
21.00
10
50
a

500
a

0.16
0.11
0.09
3 5%
4.5-5.5
4.0-6.0
3.0-7.0
00.50
04.60
07.70
10
10
a

10
a

0.04
0.13
0.12
4 4%
3.5-4.5
3.0-5.0
2.0-6.0
00.33
05.40
10.35
10
10
a

10
a

0.02
0.16
0.16
5 3%
2.5-3.5
2.0-4.0
01.01
01.91
10
10
a

0.05
0.04
6 2%
1.5-2.5
1.0-3.0
0.0-4.0
01.50
04.40
05.10
10
10
a

10
a

0.10
0.10
0.07
7 Nil
-3 - +3
-3 - +3
-4 - +4
07.30
06.40
05.10
25.
328
a

560
a

0.03
0.03
0.04
8 Nil
-3 - +3
-7 - +7
05.60
10.90
1500
140
a

0.03
0.045
Table 1. Evaluated parameters of the Nitinol wire.
a
- denotes

the number of cycles which are followed by the cumulative previous cycles.
From this diagram it is clear that for seismic mitigation, a combination of Seismic isolation
and Energy Dissipation is beneficial. Seismic Isolation can be implemented as explained
below:
- Through the reduction of the seismic response subsequent to the shift of the
fundamental period of the structure in an area of the spectrum poor in energy content
- Through the limitation of the forces transmitted to the base of the structure. A high
level of energy dissipation also characterizes this approach. So it represents a
combination of the two strategies of seismic mitigation. Isolation systems must be
capable of ensuring the following functions:
1. transmit vertical loads,
2. provide lateral flexibility,
3. provide restoring force,
4. provide significant energy dissipation.


193
In each device, the constituent elements assume one or more of the four fundamental
functions listed above. Some of the cases hybrid systems prove to be very much beneficial.
For example, the strategy need to be adopted in suspension bridges is that of isolation and
energy dissipation as the vertical cables did not provide energy dissipation characteristics.
Hence dampers need to be provided and the hybrid system provide adequate protection
during seismic response. The Table 2 below provide various energy dissipators/dampers
along with their principle of operation

Classification Principles of operation
Hysteritic devices
Yielding of metals
Friction
Visco elastic devices
Deformation of visco elastic solids
Deformation of visco elastic fluids
Fluid orificing
Re centering Devices
Fluid pressurization and orificing
Friction spring action
Phase transformation of metals (Shape
memory alloys belong to this category)
Dynamic vibration absorbers
Tuned mass dampers
Tuned liquid dampers
Table 2.Various energy dissipation devices/damper

Fig. 17. Example of Composite Rubber/SMA spring damper
The unique constitutive behaviors of Shape Memory Alloys have attracted the attention of
researchers in the civil engineering community. The collective results of these studies
suggest that they can be used effectively for vibration control of structures through vibration
isolation and energy absorption mechanisms. Possible applications of SMA based devices
on Various structures for vibration control is shown in Fig.18,
For the analysis of structures, an equivalent Single Degree Of Freedom (SDOF) system can
be utilized. The following mechanical model can represent the behavior of the energy
dissipating system (Fig.19a). If re centering device is utilized the hysterisis should be
adequately represented using mathematical models.


194

(i)

(ii)

(iii)
Fig. 18. Possible Applications of the Devices (i) Restrainers in Bridges (ii) Diagonal braces in
buildings (iii) Cable stays (iv) combination of isolators and dampers in bridges

a) b)
Fig. 19. a) Equivalent mechanical model of the SDOF scaled structure
b) The hysterisis /energy dissipation behavior of the re-centering device


195
9. Conclusion
There have been considerable research efforts in seismic response control for the past
several decades. Due to the distinctive macroscopic behaviour like super elasticity, Shape
Memory Alloys are the basis for innovative applications such as devices for protecting
buildings from structural vibrations. Super elastic properties of Nitinol wires have been
established from the experiments conducted and the salient features to be highlighted from
the study are
- The materials application can be made suitable for seismic devices like recentering,
supplementally recentering or in the case of non-recentering devices, as a variety of
hysteretic behaviors were obtained from the tests.
- Cyclic behavior of the non pre strained wires especially energy dissipation capability,
equivalent viscous damping and secant stiffness are not very sensitive to the number of
cycles in the frequency range of interest (0.5- 3Hz.) as observed from constant
amplitude loading.
- Pre-strained super elastic wires shows higher energy dissipation capability and
equivalent damping when cycled around the midpoint of the strain range obtained
from quasi-static curve. It is found from the experiment that the pre strain value of 6 %,
with amplitude cycles which covers 2-10% gives higher energy dissipation.
- For possible application of vibration control devices in structural systems, a judicious
selection of the wire under tension mode can be selected between pre strained and non-
pre strained wires. However application of pre strained wires in the system provides
excellent energy dissipation characteristics but it requires skilled and sophisticated
mechanism to maintain/provide the required pre strain.
- The mathematical model predicts the maximum energy dissipation capability of the
material namely pre strained nitinol wires under study.
- The test results shows immense promise on SMA based devices which can be used for
vibration control of variety of structures(New designs and restoration of structures).
SMA structural elements/devices can be located at key locations of the structure to
reduce the seismic vibrations.
10. Acknowledgment
The paper has been published with the kind approval of Director, CSIR-Structural
Engineering Research Centre, Chennai. The constant encouragement, and support provided
by the Director General-CSIR, Dr. Sameer K Brahmachari is gratefully acknowledged. The
help and support provided by all colleagues of Advanced Seismic Testing and Research
Laboratory in carrying out the experimental work deserve acknowledgement.
11. References
Birman, V (1997) Review of Mechanics of Shape Memory Alloy Structures Applied Mechanics
Review 50 629-645
Birman, V (1997) Effect of SMA dampers on nonlinear vibrations of elastic structures
Proceedings of SPIE 3038 ,268-76


196
Clark, P W; Aiken, I D; Kelly, J M; Higashino, M and Krumme, R C(1996) Experimental
and analytical studies of shape memory alloy damper for structural control Proc.
Passive damping (San Diego,CA 1996)
Cardone D, Dolce M, Bixio A and Nigro D 1999 Experimental tests on SMA elements
MANSIDE Project (Rome, 1999) (Italian Department for National Technical Services)
II85-104 Da GZ, Wang TM, Liu Y, Wamg CM( 2001) Surgical treatment of tibial and
femoral fractures with TiNi Shape memory alloy interlocking intra medullary nails
The international conference on Shape Memory and Superelastic Technologies and Shape
Memory materials, Kunming, China
Dolce M (1994) Passive Control of Structures Proceedings of the 10
th
European Conference
on Earthquake Engineering, Vienna, 1994.
Duerig, T; Tolomeo, D and Wholey, M. (2000), An overview of superelastic stent design
Minimally Invasive Therapy & Allied Technologies. , 2000:9(3/4) 235246.
Eaton, J P. (1999) Feasibility study of using passive SMA absorbers to minimize secondary
system structural response , Master Thesis ,Worcester polytechnic Institute, M A
Humbeeck, JV (2001) Shape Memory Alloys: a material and a technology Advanced
Engieering Materials 3 837-850
Miyazaki, S; Imai, T; Igo, Y. And Otsuka, K(1986b), Effect of cyclic deformation on the
pseudoelasticity characteristics of TiNi alloys. Metall. Trans. A. 17115120.
Pelton, A; DiCello,J; and Miyazaki,S. , (2000), Optimisation of processing and properties of
medical grade nitinol wire. Minimally Invasive Therapy & Allied Technologies. ,
2000:9(1) 107118.
Sreekala, R; Avinash, S; Gopalakrishnan, N; and Muthumani, K(2004) Energy Dissipation and
Pseudo Elasticity in NiTi Alloy Wires SERC Research Report MLP 9641/19, October 2004
Sreekala, R; Avinash, S; Gopalakrishnan, N; Sathishkumar, K and Muthumani, K(2005)
Experimental Study on a Passive Energy Dissipation Device using Shape Memory
Alloy Wires, SERC Research Report MLP 9641/21, April 2005
Sreekala,. R;. Muthumani, K; Lakshmanan, N; Gopalakrishnan, N & Sathishkumar, K
(2008),. Orthodontic arch wires for seismic risk reduction, Current Science,
ISSN:0011-3891, Vol. 95, No:11, pp 1593-1599.
Sreekala,. R,;. Muthumani,. K(2009),. Structural Application of Smart materials,. In:. Smart
Materials,. Edited by Mel Shwartz, . CRC press, Taylor & Francis Group,. pp. 4-1 to
4-7,. Taylor &Francis Publications,. ISBN-13:978-1-4200-4372-3,. Boca raton, FL,USA
Sreekala,. R;. Muthumani, K; Lakshmanan, N; Gopalakrishnan, N; Sathishkumar, K; Reddy, G,R
& Parulekar Y M. (2010),. A Study on the suitability of NiTi wires for Passive Seismic
Response Control Journal of Advanced Materials, ISSN 1070-9789, Vol. 42, No:2,pp. 65-76.
Stckel,D. and Melzer, A. ,, Materials in Clinical Applications. , ed. P. Vincentini Techna Srl.
,1995, 79198.
Wilde, K; Gardoni, P. , and Fujino Y. ,(2000) Base isolation system with shape memory alloy
device for elevated highway bridges Engineering Structures 22 222-229.
10
Whys and Wherefores of Transmissibility

N. M. M. Maia
1
, A. P. V. Urgueira
2
and R. A. B. Almeida
2

1
IDMEC-Instituto Superior Tcnico, Technical University of Lisbon
2
Faculdade de Cincias e Tecnologia, Universidade Nova de Lisboa
Portugal
1. Introduction
The present chapter draws a general overview on the concept of transmissibility and on its
potentialities, virtues, limitations and possible applications. The notion of transmissibility
has, for a long time, been limited to the single degree-of-freedom (SDOF) system; it is only
in the last ten years that the concept has evolved in a consistent manner to a generalized
definition applicable to a multiple degree-of-freedom (MDOF) system. Such a generalization
can be and has been not only developed in terms of a relation between two sets of harmonic
responses for a given loading, but also between applied harmonic forces and corresponding
reactions. Extensions to comply with random motions and random forces have also been
achieved. From the establishment of the various formulations it was possible to deduce and
understand several important properties, which allow for diverse applications that have
been envisaged, such as evaluation of unmeasured frequency response functions (FRFs),
estimation of reaction forces and detection of damage in a structure. All these aspects are
reviewed and described in a logical sequence along this chapter.
The notion of transmissibility is presented in every classic textbook on vibrations, associated
to the single degree-of-freedom system, when its basis is moving harmonically; it is defined
as the ratio between the modulus of the response amplitude and the modulus of the
imposed amplitude of motion. Its study enhances some interesting aspects, namely the fact
that beyond a certain imposed frequency there is an attenuation in the response amplitude,
compared to the input one, i.e., one enters into an isolated region of the spectrum. This
enables the design of modifications on the dynamic properties so that the system becomes
more isolated than before, as its transmissibility has decreased.
Usually, the transmissibility of forces, defined as the ratio between the modulus of the
transmitted force magnitude to the ground and the modulus of the imposed force
magnitude, is also deduced and the conclusion is that the mathematical formula of the
transmissibility of forces is exactly the same as for the transmissibility of displacements. As
it will be explained, this is not the case for multiple degree of freedom systems.
The question that arises is how to extend the idea of transmissibility to a system with N
degrees-of-freedom, i.e., how to relate a set of unknown responses to another set of known
responses, for a given set of applied forces, or how to evaluate a set of reaction forces from a
set of applied ones. Some initial attempts were given by Vakakis et al. (Paipetis & Vakakis,
1985; Vakakis, 1985; Vakakis & Paipetis, 1985; 1986), although that generalization was still


198
limited to a very particular type of N degree-of-freedom system, one where a set constituted
by a mass, stiffness and damper is repeated several times in the vertical direction. The works
of (Liu & Ewins, 1998), (Liu, 2000) and (Varoto & McConnell, 1998) also extend the initial
concept to N degrees-of-freedom systems, but again in a limited way, the former using a
definition that makes the calculations dependent on the path taken between the considered
co-ordinates involved, the latter by making the set of co-ordinates where the displacements
are known coincident to the set of applied forces.
An application where the transmissibility seems of great interest is when in field service one
cannot measure the response at some co-ordinates of the structure. If the transmissibility
could be evaluated in the laboratory or theoretically (numerically) beforehand, then by
measuring in service some responses one would be able to estimate the responses at the
inaccessible co-ordinates.
To the best knowledge of the authors, the first time that a general answer to the problem has
been given was in 1998, by (Ribeiro, 1998). Surprisingly enough, as the solution is very
simple indeed. In what follows, a chronological description of the evolution of the studies
on this subject is presented.
2. Transmissibility of motion
In this section and next sub-sections the main definitions, properties and applications will be
presented.
2.1 Fundamental formulation
The fundamental deduction (Ribeiro, 1998), based on harmonically applied forces (easy to
generalize to periodic ones), begins with the relationships between responses and forces in
terms of receptance: if one has a vector F
A
of magnitudes of the applied forces at co-
ordinates A, a vector
U
X of unknown response amplitudes at co-ordinates U and a vector
X
K
of known response amplitudes at co-ordinates K, as shown in Fig. 1.

Fig. 1. System with co-ordinates A, U, K
One may establish the following relationships:

U UA A
= X H F (1)


199

K KA A
= X H F (2)
where H
UA
and H
KA
are the receptance frequency response matrices relating co-ordinates
U and A, and K and A, respectively. Eliminating F
A
between (1) and (2), it follows that

U UA KA K
+
= X H H X (3)
or

U K
= X T X
(A)
UK
(4)
where
+
H
KA
is the pseudo-inverse of H
KA
. Thus, the transmissibility matrix is defined as:

UA KA
+
= T H H
(A)
UK
(5)
Note that the set of co-ordinates where the forces are (or may be) applied (A) need not
coincide with the set of known responses (K). The only restriction is that for the pseudo-
inverse to exist the number of K co-ordinates must be greater or equal than the number of
A co-ordinates.
An important property of the transmissibility matrix is that it does not depend on the
magnitude of the forces, one simply has to know or to choose the co-ordinates where the
forces are going to be applied (or not, as one can even choose more co-ordinates A if one is
not sure whether or not there will be some forces there and, later on, one states that those
forces are zero) and measure the necessary frequency-response-functions.
2.2 Alternative formulation
An alternative approach, developed by (Ribeiro et al., 2005) evaluates the transmissibility
matrix from the dynamic stiffness matrices, where the spatial properties (mass, stiffness,
etc.) are explicitly included.
The dynamic behaviour of an MDOF system can be described in the frequency domain by
the following equation (assuming harmonic loading):

= Z X F
(6)
where Z represents the dynamic stiffness matrix, X is the vector of the amplitudes of the
dynamic responses and F represents the vector of the amplitudes of the dynamic loads
applied to the system.
From the set of dynamic responses, as defined before, it is possible to distinguish between
two subsets of co-ordinates K and U; from the set of dynamic loads it is also possible to
distinguish between two subsets, A and B, where A is the subset where dynamic loads may
be applied and B is the set formed of the remaining co-ordinates, where dynamic loads are
never applied. One can write X and F as:
,
K A
U B

= =

X F
X F
X F
(7)
With these subsets, Eq. (6) can be partitioned accordingly:


200

AK AU K A
BK BU U B

=

Z Z X F
Z Z X F
(8)
Taking into account that co-ordinates B represent the ones where the dynamic loads are
never applied, and considering that the number of these co-ordinates is greater or equal to
the number of co-ordinates U, from Eq. (8) it is possible to obtain the unknown response
vector:

, # #
B
U BU BK K
B U =
=
+
F 0
X Z Z X
(9)
where
BU
+
Z is the pseudo-inverse of
BU
Z . Therefore, this means that the transmissibility
matrix can also be defined as

(A)
UK
=
BU BK
+
T Z Z (10)
Eq. (10) is an alternative definition of transmissibility, based on the dynamic stiffness
matrices of the structure. Therefore,

(A)
UK
+
= =
UA KA BU BK
+
T H H Z Z (11)
Taking into account that the dynamic stiffness matrix for an undamped system is described
in terms of the stiffness and mass matrices,
2
= Z K M, one can now relate the
transmissibility functions to the spatial properties of the system. To make this possible, one
must bear in mind that it is mandatory that both conditions regarding the number of co-
ordinates be valid, i. e.,
# # # # and B U K A (12)
2.3 Numerical example
An MDOF mass-spring system, presented in Fig. 2, will be used to illustrate the principal
differences observed between the transmissibilities and FRFs curves. This is a six mass-
spring system (designated as original system), possessing the characteristics described in
Table 1.
The subsets of known and unknown responses are assumed as:
{ }
2 4 6
T
T
K
X X X = X and { }
1 3 5
T
T
U
X X X = X (13)
The number of loads can be grouped in the sub-set
A
F
(even if some of them are, in certain
cases, null) and in subset
B
F
.

{ }
4 5 6
T
T
A
F F F = F and { }
1 2 3
T
T
B
F F F = F (14)
According to Eq. (11), and considering the above-defined subsets, the transmissibility matrix
is given by:


201
Original System

kg

N/m

m1
m2
m3
m4
m5
m6

k1
k2
k3
k4
k5
k6
k7
k8
k9
k10
k11

7
7
4
3
6
8

10
5

10
5
4.0x10
5
5.0x10
5
7.0x10
5
2.0x10
5
8.0x10
5
3.0x10
5
6.0x10
5
3.0x10
5
5.0x10
5

Table 1. Characteristics of the original system

Fig. 2. Mass-spring MDOF system


202

12 14 16
14 15 16 24 25 26
34 35 36 44 45 46 32 34 36
54 55 56 64 65 66
52 54 56
11 13 15
21 23 25
31 33 3

T T T
H H H H H H
T T T H H H H H H
H H H H H H
T T T
Z Z Z
Z Z Z
Z Z Z
+

=

=
(A) (A) (A)
(A) (A) (A)
(A) (A) (A)
12 14 16
22 24 26
5 32 34 36
Z Z Z
Z Z Z
Z Z Z

+
(15)
The characteristics of the system of Fig. 2 are presented in Table 1.
It may be noted from Fig. 3 that the maxima of the transmissibility curves occur all at the same
frequencies. It can also be observed that the maxima and minima of the transmissibility curves
do not coincide with the maxima and minima of the FRF curves. No simple relationships (if
any) can be established between the picks and anti-picks of the transmissibilities and FRFs.
Transmissibilities have a local nature and therefore they do not reflect the existence of the
global properties of the system (natural frequencies and damping ratios).

0 20 40 60 80 100 120 140 160 180 200
Frequency [Hz]
-280
-240
-200
-160
-120
-80
-40
0
40
80
120
M
a
g
n
i
t
u
d
e

[
d
B
]
T
14
(Orig.)
T
52
(Orig.)
T
56
(Orig.)
H
14
(Orig.)
H
35
(Orig.)

Fig. 3. Some transmissibilities and FRFs curves of the original system
2.4 Transmissibility properties
The formulation presented in section 2.1 allows us to extract some important properties for
the transmissibility matrix T
(A)
UK
. From Eq. (3) and (5) it is possible to conclude that the
transmissibility matrix is independent from the force vector
A
F (Note that
A
F is eliminated
between eqs. (1) and (2)). This means that any change verified in one of the force values,
acting along with co-ordinates of set A, will not affect T
(A)
UK
. This change can be due, for
instance, to the alteration of mass values associated to co-ordinates A or stiffness values of
springs interconnecting those co-ordinates.
Additionally, to highlight that characteristic of matrix T
(A)
UK
, it can be verified from Eq. (10)
that there is no part of matrix Z involving co-ordinates of set A (neither
AK
Z nor
AU
Z ). This
statement reinforces the previous conclusion extracted from Eq. (5) and will lead to the
formulation of two properties, as follows:


203
Property 1. The transmissibility matrix does not change if some modification is made on the mass
values of the system where the loads can be applied subset A.
Property 2. The transmissibility matrix does not change if some modification is made on the stiffness
values of springs interconnecting co-ordinates of subset A (where the loads can be applied).
In fact, any changes in the mass values associated to co-ordinates A and/or any changes in
the stiffness values of springs interconnecting co-ordinates A, will affect the inertia forces
and elastic forces, respectively, acting along those co-ordinates and thus belonging to
A
F .

The same MDOF system presented in Fig. 2 will be used to illustrate the transmissibility
properties. In Table 2 four different modifications are made in the original system.
Situations I and II correspond to modifications on the original masses; situations III and IV
correspond to modifications on stiffness. Situations I and III only involve co-ordinates A,
whereas situations II and IV involve co-ordinates pertaining to both sets A and B.
Choosing, for instance, the transmissibility function
( )
52
A
T , one obtains the results presented
in Figs. 4 and 5, where one can see that
( )
52
A
T and
( )
14
A
T remain the same only when changes
are made at co-ordinates A, where the forces are applied.
2.5 Evaluation of the transmissibility from measurement responses
In 1999, (Ribeiro et al., 1999) and (Maia et al., 1999) showed how the transmissibility matrix
could be evaluated directly from the measurement of the responses, rather than measuring
the frequency response functions. In Eq. (4), the problem is to evaluate the U K values of
T
(A)
UK
knowing X
U
and
K
X . This can be achieved by applying various sets of forces, at a
time, on co-ordinates A. Let
(1)
A
F be the first set of applied forces (amplitudes). Then,

= X T X
(1) (A) (1)
U UK K
(16)

Original System Situation I Situation II Situation III Situation IV

kg

N/m

m1
m2
m3
m4
m5
m6

k1
k2
k3
k4
k5
k6
k7
k8
k9
k10
k11

7
7
4
3
6
8

10
5

10
5
4.0x10
5
5.0x10
5
7.0x10
5
2.0x10
5
8.0x10
5
3.0x10
5
6.0x10
5
3.0x10
5
5.0x10
5

--
--
--
13
12
--
--
--
--
--
--
--
--
--
--
--
--
--

14
--
--
--
--
13
--
--
--
--
--
--
--
--
--
--
--
--

14
--
--
--
--
13
--
--
--
--
--
--
--
--
--
--
9.0x10
5
9.0x10
6

--
--
--
--
--
--
--
--
--
--
10.0x10
5
--
--
--
--
--
--
--

-- unchanged value
Table 2. Characteristics of the modifications made in the original system


204
0 20 40 60 80 100 120 140 160 180 200
Frequency [Hz]
-80
-60
-40
-20
0
20
40
60
80
100
T
r
a
n
s
m
i
s
s
i
b
i
l
i
t
y

[
d
B
]
T
52
(orig.)
T
52
(Sit. I)
T
52
(Sit. II)

Fig. 4. Transmissibility
( )
52
A
T , for the original system and the modified systems I e II

0 20 40 60 80 100 120 140 160 180 200
Frequency [Hz]
-100
-80
-60
-40
-20
0
20
40
60
80
T
r
a
n
s
m
i
s
s
i
b
i
l
i
t
y

[
d
B
]
T
14
(orig.)
T
14
(Sit. III)
T
14
(Sit. IV)

Fig. 5. Transmissibility
( )
14
A
T , for the original system and the modified systems III e IV
However, when forces change, the transmissibility matrix does not, although the responses
themselves do. Therefore, if another test is performed, with a set of forces linearly
independent of the first one though applied at the same co-ordinates
(2)
A
F , a new set of U
equations can be obtained, which are linearly independent of the first ones. If K tests are
undertaken on the structure, with linearly independent forces always applied at the A set of
co-ordinates, one can obtain a system of U K equations to solve for the same number of
(A)
UK
T unknowns:

=

" " X X X T X X X
(1) (2) (K) (A) (1) (2) (K)
U U U UK K K K
(17)
From which

1

=

" " T X X X X X X
(A) (1) (2) (K) (1) (2) (K)
UK U U U K K K
(18)


205
Note: an easy way to obtain the K sets of linearly independent forces is to apply a single
force at each of the A locations at a time.
2.6 The distributed forces case
In (Ribeiro, Maia, & Silva, 2000b) the authors discussed the transmissibility between the two
sets of co-ordinates U and K when a distributed force is applied to the structure. Such a force
should be discretized to form the A set of applied forces. The issue was then to study what
happened if one only took a subset of those co-ordinates A, which implies the reduction or
condensation of the applied forces to such a subset and to ask the question: how to study
the transmissibility of responses from a set of condensed forces?. Let the set A be composed
by the set C to where one wishes to condense the forces and the set D of the remaining co-
ordinates, so that:

=

F
F
F
C
A
D
(19)
If one wishes to condense F
A
to F
C
, one needs to assume some relationship between F
D

and F
C
, i.e, one cannot contemplate the case where all the applied forces are completely
independent from each other. However, this is not a big restriction, as it seems reasonable to
expect that the applied forces exhibit a more or less fixed spatial pattern along the structure.
Therefore, let us assume a linear relationship between the sets of forces F
D
and F
C
,
through the matrix P
DC
:

= F P F
D DC C (20)
If matrices
UA
H and
KA
H from eqs. (1) and (2) are partitioned into
[ ]
UA UC UD
= # H H H (21)

[ ]
KA KC KD
= # H H H
(22)
and one has Eq. (19) into account, then eqs. (1) and (2) become

U UC UD
= + X H F H F
C D
(23)

K KC KD
= + X H F H F
C D
(24)
Substituting Eq. (20) in eqs. (23) and (24) and eliminating
C
F , it follows that
( )( )
U UC UD KC KD K
+
= + + X H H P H H P X
DC DC
(25)
Therefore, one has now, instead of Eq. (4), another one relating
U
X and
K
X , through a new
transmissibility matrix referred to the new subset of co-ordinates C:

U K
= X T X
(C)
UK
(26)


206
where
( )( )
UC UD KC KD
+
= + + T H H P H H P
(C)
UK DC DC
(27)
Note: for the pseudo-inverse to exist, the number of K co-ordinates must be higher than the
number of C co-ordinates. This is obviously verified, as # # K A and # # A C > . It should
also be stressed that although referred to a reduced set of co-ordinates C where the forces
are applied, the transmissibility matrix still relates the responses U and K and so it keeps the
same size.
In (Ribeiro, Maia, & Silva, 2000a), (Ribeiro, Maia, & Silva, 2000b) and (Maia et al., 2001) the
authors summarize some of the previous works and suggest some other possible
applications for the transmissibility concept, namely in the area of damage detection. In this
area, there has been some activity trying to use the transmissibility as defined, as well as
some other variations of it, with limited results in [(Sampaio et al., 1999; 2000; 2001)], but
with some promising evolution in [(Maia et al., 2007)].
An example is presented in order to illustrate the above discussion: a cantilevered beam is
subjected to the loading shown in Fig. 6.

Fig. 6. Example of a loaded beam
It is assumed that forces are applied at co-ordinates 1 to 6 but the forces at co-ordinates 2 to
5 can be related to those applied at 1 and 6 through the expression:

2
3 1
4 6
5
4 1
3 2
1

2 3 5
1 4
f
f f
f f
f

= =

F P F
D DC C
(28)
One can further assume that the responses at co-ordinates 1 and 2 can be measured and
those at co-ordinates 4 and 5 can be computed through the transmissibility, i.e.,
U K
= X T X
(C)
UK
, with
4
5
U

=

X
X
X
,
1
2
K

=

X
X
X
and


207

( )( )
41 46 42 43 44 45 11 16
51 56 52 53 54 55 21 26

4 1
3 2
1
2 3 5
1 4
UC UD KC KD
H H H H H H H H
H H H H H H H H
+
= + +

= +

T H H P H H P
T
(C)
UK DC DC
(C)
UK

12 13 14 15
22 23 24 25
4 1
3 2
1
2 3 5
1 4
H H H H
H H H H
+

(29)
2.7 The random forces case
Often one has to deal with random forces, for instance when a structure is submitted to
environmental loads. The cases that have been addressed so far were limited to harmonic or
periodic forces. The generalization to random forces has been derived in [(Ribeiro et al.,
2002; Fontul et al., 2004)], now in terms of power spectral densities, rather than in terms of
response amplitudes. Let
KK
S denote the auto-spectral density of the responses
K
X and
KU
S the cross-spectral density between responses
K
X and
U
X . Then, it can be shown (see
[(Fontul, 2005)] for specific details) that both are related through the same transmissibility
matrix as before (using Eq. (5), for instance):

T T
KU KK
= S T S
(A)
UK
(30)
2.8 Some possible applications
2.8.1 Transmissibility of motion in structural coupling
This topic has been addressed in (Devriendt, 2004; Ribeiro et al., 2004; Devriendt & Fontul,
2005). Let us consider a main structure, to which an additional structure is coupled though
some coupling co-ordinates, i.e., the additional structure applies a set of forces (and moments)
to the main structure. As the transmissibility between two sets of responses on the main
structure does not depend on the magnitude of those forces, the transmissibility matrix of the
main structure is equal to the transmissibility matrix of the total structure (main + additional).
In other words, under certain conditions, the transmissibility matrix of the main structure
remains unchanged, even if an additional structure is coupled to the main one. To make this
property valid it is necessary to consider a sufficient number of coupled co-ordinates.
Although it might be argued that a reduced number of coupling co-ordinates would hamper
the results since it would not include information about some modes, it has been shown in
(Devriendt, 2004; Ribeiro et al., 2004; Devriendt & Fontul, 2005) that, as long as there is enough
information regarding the modes included in the frequency range of interest the minimum
number of coupling co-ordinates can be reduced without deterioration of the results.
2.8.2 Evaluation of unmeasured frequency response functions
Recent papers [(Maia et al., 2008; Urgueira et al., 2008)] have explored some invariance
properties of the transmissibility, namely when modifications are made in terms of masses


208
and/or stiffnesses at the co-ordinates where the forces are applied, to be able to estimate the
new FRFs in locations that become no longer accessible. For instance, if one calculates the
transmissibility matrix at some stage between two sets of responses for a given set of
applied forces and later on there are some modifications at the force co-ordinates due to
some added masses, the FRFs will change but the transmissibility remains the same. This
allows the estimation of the new FRFs. So, initially one has
(1) (1) (1) (1) +
=
U UA KA K
X H H X and
later on one has
(2) (2) (2) (2)
U UA KA K
+
= X H H X . As the transmissibility remains unchanged, one
has

(1) (1) (2) (2)
UA KA UA KA
+ +
= = T H H H H
(A)
UK
(31)
and one can calculate, for instance,
(2)
UA
H , given by:

(2) (2)
UA KA
= H T H
(A)
UK
(32)
2.9 Direct transmissibility
From the definition given before one has,

U UA KA K K
+
= = X H H X T X
(A)
UK
(33)
which, as explained, is a generalisation from the one degree of freedom system. However, in
some cases it might be useful to divide two responses directly. In strict sense that is a
transmissibility only if a single force is applied. Otherwise, one has to name it differently,
like pseudo-transmissibility (e.g. (Sampaio et al., 1999)), scalar transmissibility (Devriendt et
al., 2010) or direct transmissibility, which is the one we shall adopt.
Direct transmissibilities will depend on the force magnitudes (as well as location, of course).
For example, dividing Eq. (33) by one of the amplitudes
K
X , say
s
X , one has:

or
U s K s Us Ks
X X = = T T
(A) (A)
UK UK
X X
(34)
It is easier to understand the implications of both definitions through an example: let
U
X ,
K
X , and
A
F be given respectively by

1 3 5
2 4 6
U K A
X X F
X X F

= = =

X X F (35)
The relation between
U
X and
K
X would be:

1 13 14 3 1 13 3 14 4
2 23 34 4 2 23 3 24 4

X T T X X T X T X
X T T X X T X T X
= +
=

= +

(36)
Dividing Eq. (36) by, say,
3
X , it follows that
or
1 3 13 14 4 3 13 13 14 43
2 3 23 24 4 3 23 23 24 43

X X T T X X T T
X X T T X X T T

= + = +
= + = +
(37)


209
One can also write:

1 15 5 16 6
15 5 16 6 1
13
3 35 5 36 6 3 35 5 36 6

X H F H F
H F H F X
X H F H F X H F H F
= + +
= =
= + +
(38)
From Eq. (38) it is clear that the direct transmissibility
13

depends on the magnitudes of
5
F
and
6
F , unless the relation
5 6
F F remains constant. Only in the case where there is just a
single force one has a coincidence between both types of transmissibility.
Both kinds of definitions can be useful. For instance, concerning now the direct
transmissibilities, one can see that from Eq. (37) one can calculate
13
and
23
from
43
, and
one can eliminate
43
between both equations and establish a relationship between
13
and
23
, therefore allowing the evaluation of one of them from the other. Moreover and similarly
to what was mentioned in section 2.5, the direct transmissibilities allow the calculation of
the other ones.
To illustrate the main differences between the curves of general and the direct
transmissibilities, the MDOF system presented in Fig. 2 has been used. In Fig. 7 some direct
transmissibility curves are presented.

0 20 40 60 80 100 120 140 160 180 200
Frequency [Hz]
-80
-60
-40
-20
0
20
40
60
80
D
i
r
e
c
t

T
r
a
n
s
.

[
d
B
]
23
(Orig.)
45
(Orig.)
56
(Orig.)

Fig. 7. Some direct transmissibility curves from the system of Fig. 2
By comparing the results of the transmissibilities of Fig. 3 with the curves obtained with the
direct transmissibilities, Fig. 7, one can see that both look like FRFs, though it may be noted
that in the case of the transmissibilities all the maxima occur at the same frequencies; the
same is not true with the direct transmissibilities, where each curve presents distinct
maxima.
2.10 Other applications
Other works have presented the possibility of using the transmissibility concept for model
updating (Steenackers et al., 2007) and to identify the dynamic properties of a structure
(Devriendt & Guillaume, 2007; Devriendt, De Sitter, et al., 2009; Devriendt et al., 2010).


210
Other recent studies have applied the transmissibility to the problem of transfer path
analysis in vibro-acoustics (Tcherniak & Schuhmacher, 2009) and for damage detection
(Canales et al., 2009; Devriendt, Vanbrabant, et al., 2009; Urgueira et al., 2011).
3. Transmissibility of forces
3.1 In terms of frequency response functions
Another important topic may be the prediction of the dynamic forces transmitted to the
ground when a machine is working. For a single degree of freedom, the solution is well
known and the transmissibility is defined as the ratio between the transmitted load (the
ground reaction) and the applied one, for harmonic excitation. For an MDOF system, one
has to relate the known applied loads (
K
F ) to the unknown reactions (
U
F ), Fig. 8.
The displacements at the co-ordinates of one set (the set of the reactions) are constrained, so
they must also be known (possibly zero).
The inverse problem may also be of interest, i.e., to estimate the loads applied to a structure
(wind, traffic, earthquakes, etc.) from the measured reaction loads. Once the load
transmissibility matrix is established between the appropriate sets, the measurement of the
reactions is expected to allow for the estimation of the external loads.

1
K
F
2
K
F
3
K
F
[ ] [ ] [ ] , , M K C
1
K
F
2
K
F
3
K
F
[ ] [ ] [ ] , , M K C
1
U
F
2
U
F
3
U
F

Fig. 8. Structure with applied loads and reactions in dynamic equilibrium
This topic has been addressed in (Maia et al., 2006); the force transmissibility may also be
defined either in terms of FRFs or in terms of dynamic stiffnesses. Let
K
X and
U
X be the
responses corresponding to
K
F and
U
F , respectively, and
C
X the responses at the
remaining co-ordinates; then,

K KK KU
K
U UK UU
U
C CK CU

=

X
F
X
F
X

(39)
Assuming the responses at the reactions co-ordinates as zero, i.e.,
U
= X 0 , it follows that:

1
U UK K UU
= F F
(40)


211
Therefore, the force transmissibility is defined as:

1
UK UK UU
= T (41)
If the displacements at the co-ordinates of the reactions are not zero, or in the more general
case when the two sets of loads are not the applied loads and the reactions, but any disjoint
sets that encompass all the loads applied to the structure, it is easy to show (Maia et al.,
2006) that:

1
U UK K U UU
= + F F X (42)
3.2 In terms of dynamic stiffness
Instead of Eq. (39) one has now:

K
K KK KU KC
U
U UK UU UC
C

=

X
F
X
F
X

(43)
Assuming fictitious loads
C
F at the remaining co-ordinates and rearranging, one obtains:

K KK KC KU K
C CK CC CU C
U UK UC UU U

=

F X
F X
F X

(44)
Defining { }
T
E K C
= X X X and { }
T
E K C
= F F F and assuming, as before, that at the
reaction co-ordinates there is no motion (
U
= X 0 ), one can write:

E EE E
U UE E
=
=
F X
F X
(45)
Eliminating
E
X
between eqs.(45), it follows that

1
U UE E EE
= F F
(46)
and the force transmissibility becomes now:

1
UE UE EE
= T
(47)
Note that because { }
T
C E K
= = F 0 , F F 0 , and thus only the columns of
UE
T corresponding
to
K
F are relevant to the transmissibility between the two sets of loads, the sub-matrix
UK
T .
One should also note that, in contrast with the SDOF system, the transmissibility of forces is
different from the transmissibility of displacements.
Simply to illustrate the application of the concept, a numerical example is presented. The
model is shown in Fig. 9, similar to the one of Fig. 3, where the displacements at co-
ordinates 1 and 2 are now zero, i.e.,
1 2
0 = = X X . External forces are applied at co-ordinates
5 and 6 and the reactions happen at co-ordinates 1 and 2.


212

Fig. 9. Structure model in study
The force transmissibility between the two sets of loads forces at 5 and 6 being known (set
K) and forces at 1 and 2 being unknown (set U) was computed using both described
methods.

5
K
6
F
F

=

F ,
1
2
U
F
F

=

F ,
5
6
0
0
E
F
F

=

F ,
5
6
K
X
X

=

X ,
1
2
U
X
X

=

X ,
5
6
3
4
E
X
X
X
X

=

X (48)
Equation (41) becomes:

11 12
21 22
1
11 12 15 16
21 22 25 26
H H
UK
H H
T T
H H H H
H H H H T T

= =

T
(49)
where the subscript H means that the transmissibility has been computed using FRFs.
Equation (47) becomes:

11 12 13 14
21 22 23 24
1
55 56 53 54
15 16 13 14 65 66 63 64
25 26 23 24 35 36 33 34
45 46 43 44
Z Z Z Z
UE
Z Z Z Z
Z Z Z Z
T T T T
Z Z Z Z Z Z Z Z
Z Z Z Z Z Z Z Z T T T T
Z Z Z Z

= =

T
(50)
from which


213

11 12
21 22
Z Z
UK
Z Z
T T
T T

=

T
(51)
where the subscript Z means that the transmissibility has been computed using dynamic
stiffness matrices. The results obtained by using equations (49) and (51) superimpose
perfectly, as expected. Two of the four transmissibilities are presented in Fig. 10 to illustrate
this fact.

0 20 40 60 80 100 120 140 160 180 200
Frequency [Hz]
-80
-60
-40
-20
0
20
40
60
F
o
r
c
e

T
r
a
n
s
.

[
d
B
]
H
11

Z
11

0 20 40 60 80 100 120 140 160 180 200
Frequency [Hz]
-80
-60
-40
-20
0
20
40
60
F
o
r
c
e

T
r
a
n
s
.

[
d
B
]
H
12

Z
12

Fig. 10. Comparison between corresponding force transmissibility terms computed from
FRFs (
11 22
and
H H
T T ) and dynamic stiffness matrices (
11 22
and
Z Z
T T ).
It may be noted from Fig. 10 that the maxima of the force transmissibility curves also occur
all at the same frequencies.
4. Conclusions
The transmissibility concept for multiple degree-of-freedom systems has been developed
and applied for the last ten years and the interest in this matter is continuously growing. In
this paper a general overview has been given, concerning the main achievements so far and
it has been shown that the various ways in which transmissibility can be defined and
applied opens various possibilities for research in different domains, like system
identification, structural modification, coupling analysis, damage detection, model
updating, vibro-acoustic applications, isolation and vibration attenuation.
5. Acknowledgment
The authors greatly appreciate the financial support of FCT, under the research program
POCI 2010.
6. References
Canales, G., Mevel, L., Basseville, M. (2009). Transmissibility Based Damage Detection,
Proceedings of the 27th International Operational Modal Analysis Conference (IMAC
XXVII), Orlando, Florida, U.S.A.


214
Devriendt, C.M.G., Fontul, M. (2005). Transmissibility in Structural Coupling: Identification
of the Sufficient Set of Coupling Forces, Proceedings of the 1st International
Operational Modal Analysis Conference (IOMAC 1), Copenhagen, Denmark.
Devriendt, C., Vanbrabant, K., De Sitter, G., Presznzk, F., De Troyer, T., Deckers, K.,
Steenackers, G., Vanlanduit, S., Guillaume, P. (2009). Structural Health Monitoring
in Changing Operational Conditions Using Transmissibility measurements,
Proceedings of International Conference on Engineering Dynamic (ICEDyn 2009),
Ericeira, Portugal.
Devriendt, C. (2004). Transmissibility on Coupled Structures (Final Project Report). Instituto
Superior Tcnico, Portugal.
Devriendt, C., De Sitter, G., Guillaume, P. (2010). An Operational Modal Analysis Approach
Based on Parametrically Identified Multivariable Transmissibilities, Mechanical
Systems and Signal Processing, Vol. 24, No. 5, pp. 1250-1259, ISSN: 0888-3270.
Devriendt, C., De Sitter, G., Vanlanduit, S., Guillaume, P. (2009). Operational Modal
Analysis in the Presence of Harmonic Excitations by the Use of Transmissibility
Measurements, Mechanical Systems and Signal Processing, Vol. 23, No. 3, pp. 621-635,
ISSN:0888-3270.
Devriendt, C., Guillaume, P. (2007). The use of Transmissibility Measurements in Output-
Only Modal Analysis, Mechanical Systems and Signal Processing, Vol. 21, No. 7, pp.
2689-2696, ISSN: 0888-3270.
Fontul, M. (2005). Transmissibility in Multi-Degree-of-Freedom Systems (Ph. D Thesis).
Fontul, M., Ribeiro, A.M.R., Silva, J.M.M., Maia, N.M.M. (2004). Transmissibility Matrix in
Harmonic and Random Processes, Shock and Vibration, Vol. 11, No. 5-6, pp. 563572,
ISSN: 1070-9622.
Liu, W. (2000). Structural Dynamic and Testing of Coupled Structures (Ph. D Thesis).
Liu, W., Ewins, D. J. (1998). Transmissibility Properties of MDOF Systems, Proceedings of the
16th International Operational Modal Analysis Conference (IMAC XVI), Santa Barbara,
California, pp. 847-854.
Maia, N.M.M., Almeida, R.A.B., Urgueira, A.P.V. (2008). Understanding Transmissibility
Properties, Proceedings of the 26th International Modal Analysis Conference (IMAC
XXVI), ISBN: 0-912053-98-4, Orlando, Florida, U.S.A.
Maia, N.M.M., Fontul, M., Ribeiro, A.M.R. (2006). Transmissibility of Forces in Multiple-
Degree-of Freedom Systems, Proceedings of International Conference on Noise and
Vibration Engineering (ISMA 2006), Leuven, Belgium.
Maia, N.M.M., Ribeiro, A.M.R., Fontul, M., Montalvo, D., Sampaio, R.P.C. (2007). Using the
Detection and Relative Damage Quantification Indicator (DRQ) with
Transmissibility, Key Engineering Materials, Vol. 347, pp. 455-460, ISSN: 1662-9795.
Maia, N.M.M., Silva, J.M.M., Ribeiro, A.M.R. (1999). The Transmissibility Concept in Multi-
Degree-of-Freedom Proceedings of Systems, Inverse Methods in Structural Dynamics,
European Mechanics Colloquium, EUROMEC 401, Liverpool, UK.
Maia, N.M.M., Silva, J.M.M., Ribeiro, A.M.R. (2001). The Transmissibility Concept in Multi-
Degrees-of-Freedom Systems, Mechanical Systems and Signal Processing, Vol. 15, No.
1, pp. 129-137, ISSN: 0888-3270.
Paipetis, S.A., Vakakis, A.F. (1985). A Method of Analysis for Unidirectional Vibration
Isolators with Many Degrees of Freedom, Journal of Sound and Vibration, Vol. 98, No.
1, pp. 13-23, ISSN: 0022-460X.


215
Ribeiro, A.M.R., Fontul, M., Silva, J.M.M., Maia, N.M.M. (2002). Transmissibility Matrix in
Harmonic and Random Processes, Proceedings of the International Conference on
Structural Dynamics Modelling (SDM), Funchal, Madeira, Portugal.
Ribeiro, A.M.R., Silva, J.M.M., Maia, N.M.M., Fontul, M. (2004). Transmissibility in
Structural Coupling, Proceedings of International Conference on Noise and Vibration
Engineering (ISMA 2004), Leuven, Belgium.
Ribeiro, A.M.R., Maia, N.M.M., Silva, J.M.M. (2000a). On the Generalisation of the
Transmissibility Concept, Mechanical Systems and Signal Processing, Vol. 14, No. 1,
pp. 29-35, ISSN: 0888-3270.
Ribeiro, A.M.R. (1998). On the Generalization of the Transmissibility Concept, Proceedings of
the NATO/ASI Conference on Modal Analysis and Testing, Sesimbra, Portugal.
Ribeiro, A.M.R., Fontul, M., Maia, N.M.M., Silva, J.M.M. (2005). Further Developments on
the Transmissibility Concept for Multiple Degree of Freedom Systems, Scientific
Bulletin of the Politehnica University of Timisoara, Transactions on Mechanics, Vol. 50,
No. 64, pp. 12246077, Timisoara, Romania.
Ribeiro, A.M.R., Maia, N.M.M., Silva, J.M.M. (1999). Experimental Evaluation of the
Transmissibility Matrix, Proceedings of the 17th International Modal Analysis
Conference (IMAC XVII), Kissimmee FL, USA, pp. 1126-1129, ISBN: 0-912053-64-X.
Ribeiro, A.M.R., Maia, N.M.M., Silva, J.M.M. (2000b). Response Prediction From a Reduced
Set of Known Responses Using the Transmissibility, Proceedings of. the 18th
International Modal Analysis Conference (IMAC XVIII), pp. 425-427, ISBN: 0-912053-
67-4, San Antonio TX, USA.
Sampaio, R.P.C., Maia, N.M.M., Ribeiro, A.M.R., Silva, J.M.M. (1999). Damage Detection
Using the Transmissibility Concept, Proceedings of the Sixth International Congress on
Sound and Vibration (ICSV 6), Copenhagen, Denmark.
Sampaio, R.P.C., Maia, N.M.M., Ribeiro, A.M.R., Silva, J.M.M. (2001). Transmissibility
Techniques for Damage Detection, Proceedings of 19th International Modal Analysis
Conference (IMAC XIX), Kissimmee, Florida, USA, pp. 1524-1527.
Sampaio, R.P.C., Maia, N.M.M., Silva, J.M.M., Ribeiro, A.M.R. (2000). On the Use of
Transmissibility for Damage Detection and Location, Proceedings of the European
COST F3 Conference on System Identification & Structural Health Monitoring,
Polytechnic University of Madrid, Spain, pp. 363-376.
Steenackers, G., Devriendt, C., Guillaume, P. (2007). On the Use of Transmissibility
Measurements for Finite Element Model Updating, Journal of Sound and Vibration,
Vol. 303, No. 3-5, pp. 707-722, ISSN: 0022-460X.
Tcherniak, D., Schuhmacher, A.P. (2009). Application of Transmissibility Matrix Method to
NVH Source Contribution Analysis, Proceedings of the 27th International Modal
Analysis Conference (IMAC XXVII), Orlando, Florida, U.S.A.
Urgueira, A.P.V., Almeida, R.A.B., Maia, N.M.M. (2008). Experimental Estimation of FRFs
Using the Transmissibility Concept, Proceedings of International Conference on Modal
Analysis Noise and Vibration Engineering (ISMA 2008), ISBN: 978-90-73802-86-5,
Leuven, Belgium.
Urgueira, A.P.V., Almeida, R.A.B., Maia, N.M.M. (2011). On the use of the transmissibility
concept for the evaluation of frequency response functions, Mechanical Systems and
Signal Processing, Vol. 25, No. 3, ISSN: 0888-3270, 940-951.


216
Vakakis, A.F. (1985). Dynamic Analysis of a Unidirectional Periodic Isolator, Consisting of
Identical Masses and Intermediate Distributed Resilient Blocks. Journal of Sound and
Vibration, Vol. 103, No. 1, pp. 25-33, ISSN: 0022-460X.
Vakakis, A.F., Paipetis, S.A. (1985). Transient Response of Unidirectional Vibration Isolators
with Many Degrees of Freedom, Journal of Sound and Vibration, Vol. 99, No. 4, pp.
557-562, ISSN: 0022-460X.
Vakakis, A.F., Paipetis, S.A. (1986). The Effect of a Viscously Damped Dynamic Absorber on
a Linear Multi-Degree-of-Freedom System, Journal of Sound and Vibration, Vol. 105,
No. 1, pp. 49-60, ISSN: 0022-460X.
Varoto, P.S., McConnell, K.G. (1998). Single Point vs Multi Point Acceleration
Transmissibility Concepts in Vibration Testing, Proceedings of the 12th International
Modal Analysis Conference (IMAC XVI), pp. 83-90, ISBN 0-912053-59-3, Santa
Barbara, California, USA.
0
Control Design Methodologies for Vibration
Mitigation on Wind Turbine Systems
Ragnar Eide and Hamid Reza Karimi
Department of Engineering, Faculty of Engineering and Science University of Agder
Postboks Grimstad
Norway
1. Introduction
The worlds energy consumption from the beginning of the industrial revolution in the 18
th. century and until today has increased at a tremendous degree. Since a large part of the
energy has come from sources like oil and coal have the negative impacts on the environment
increased proportionally. Therefore, more sustainable and climate friendly energy production
methods are emphasized among researchers and environmentalists throughout the world.
This is the reason why renewable energies, and wind power particularly, have now become
an essential part of the energy programs for most of governments all over the world. One
example is seen by the outcome of the European Conference for Renewable Energy in Berlin
in 2007 where EU countries dened ambitious goals when it comes to the increase in use of
renewable energy resources. One of the goals was that by 2020, the EU would seek to get 20%
of energy consumption from renewable energies.
Wind power, in conjunction with other renewable power production methods, has been
suggested to play a more and more important role in the future power supply (Waltz, 2008)
and (Lee & Kim, 2010). One of the reasons for these expectations is the enormous available
potential when it comes to wind resources. One of the most comprehensive study on this
topic (Archer & Jacobson, 2005) found the potential of wind power on land and near-shore to
be 72TW, which alone could have provided over ve times the worlds current energy use in
all forms averaged over a year.
The World Wind Energy Association (WWEA) estimates the wind power investment
worldwide to expand fromapproximately 160 GWinstalled capacity at the end of 2010 to 1900
GW installed capacity by 2020. One example is from the USA, where the current contribution
of electricity from wind power is merely 1,8% in (2009). However, the U.S Department of
Energy is now laying a framework to get as much as 20% contribution by the year 2030.
Due to the economical advantages of installing larger wind turbines (WTs), the typical size
of utility-scale turbines has grown dramatically over the last three decades. In addition to
the increasing turbine-sizes, cost reduction demands imply use of lighter and hence more
exible structures. If the energy-price from WTs in the coming years are to be competitive
with other power production methods, an optimal balance must be made between maximum
power capture on one side, and load-reduction capability on the other side. To be able to
obtain this is a well dened control-design needed to improve energy capture and reduce
11
dynamic loads. This combined with the fact that maintenance and constant supervision of
WTs at offshore locations is expensive and very difcult, which has further increased the need
of a reliable control system for fatigue and load reduction. New advanced control approaches
must be designed such as to achieve to the 20- to 25-year operational life required by todays
machines (Wright, 2004).
In this paper an above rated wind speed (Region III) regulation of a Horizontal Axis Wind
Turbine (HAWT) is presented. The rst method is Disturbance Accommodating Control
(DAC) which is compared to the LQG controller. The main focus in this work is to use these
control techniques to reduce the torque variations by using speed control with collective blade
pitch adjustments. Simulation results show effects of the control methodologies for vibration
mitigation on wind turbine systems.
The paper is organized as follows. Section 2 will be devoted to the modeling phase. The aim
is to come up with a simplied state space model of the WT appropriate to be used in the
control design in the subsequent sections. The control system design is covered in Section 3.
After a historical overview and a state-of-the-art presentation of WT control are the elements
involved in the practical control designed merged with a theoretical description of each topic.
The model will then be implemented into a simulation model in the in the MATLAB/Simulink
environment in Section 4. Finally, the conclusions and further improvement suggestions are
drawn in Section 5.
2. Modeling of the wind turbine
2.1 Introduction
There are different methods available for modeling purposes. Large multi-body dynamics
codes, as reported in (Elliot & Wright, 2004), divide the structure into numerous rigid body
masses and connect these parts with springs and dampers. This approach leads to dynamic
models with hundreds or thousands of DOFs. Hence, the order of these models must
be greatly reduced to make them practical for control design (Wright & Fingersh, 2008).
Another approach is an assumed modes method. This method discretize the WT structure
such that the most important turbine dynamics can be modeled with just a few degrees of
freedom. Designing controllers based on these models is much simpler, and captures the
most important turbine dynamics, leading to a stable closed-loop system (Wright & Fingersh,
2008). The method is for instance used in FAST, a popular simulation program for design and
simulation of control system (Jonkman & Buhl, 2005).
This section presents a simplied control-oriented model. In this approach, a state space
representation of the dynamic systemis derived fromof a quite simple mechanical description
of the WT. This state space model is totally non-linear due to the aerodynamics involved,
and will thus be linearized around a specic operation point. As reported in (Wright &
Fingersh, 2008) with corresponding references, good results are obtained by using linearized
time invariant models for the control design.
When modeling a WT one may need to combine different models, each representing
interacting subsystems, as Figure 1 below depicts. Here we can see how the WT is simplied
to consist of the aerodynamic-, mechanical- (drive train), and electric subsystem, and that the
blade pitch angle reference and the power reference in this case are controllable inputs.
The WT is a complicated mechanical system with many interconnecting DOF. However, some
of the couplings are rather weak and can be neglected (Ekelund, 1997). For instance, the
Control Design Methodologies for Vibration Mitigation on Wind Turbine Systems 3
Fig. 1. WT subsystems with corresponding models
connection between the dynamics of the transmission and the tower is neglected in modeling
of the mechanical system. The dynamics of the generator and the electrical system are also
neglected by regarding the reaction torque from the generator as a xed value. Also when
considering the wind, the approach is to model the wind as simply a scalar input affecting the
rotor state.
2.2 State space representation
The nonlinearities of a WT system, for instance due to the aerodynamics, may bring along
challenges when it comes to the control design. Since the control input gains of a pitch control
usually is the partial derivative of the rotor aerodynamic torque with respect to blade pitch
angle variations, these input gains will depend on the operating condition, described by a
specic wind and rotor speed. A controller designed for a turbine at one operating point
may give poor results at other operating conditions. In fact, a controller which has shown
to stabilize the plant for a limited range of operation points, may cause unstable closed-loop
behavior in other conditions.
A method which bypasses the challenges of directly involving the nonlinear equations is by
using a linear time invariant system(LTI) on state space form. Such a systemrelates the control
input vector u and output of the plant y using rst-order vector ordinary differential equation
on the form
x = Ax + Bu +u
D
y = Cx + Du
(1)
where x is the system states and matrices A, B, , C and D are the state-, input-, disturbance-,
output- and feedthrough- matrix, respectively, and u
D
is the disturbance input vector.
The representation of the system as given in Eqns. (1) has many advantages. Firstly, it allows
the control designer to study more general models (i.e., not just linear or stationary ODEs).
Having the ODEs in state variable formgives a compact, standard formfor the control design.
State space systems also contain a description of the internal states of the system along with
the input-output relationship. This helps the control designer to keep track of all the modes
219 Control Design Methodologies for Vibration Mitigation on Wind Turbine Systems
(which is important to do since a systemcan be internally unstable, although it is input-output
stable). As shall be shown in the following is a description of the dynamics on state-space a
good starting point for the further controller and observer designs.
The WT drive-train modeled with its high and low speed shaft separated by a gearbox is
shown in Figure 2. As its seen, the drive train is modeled as a simple spring-damper
conguration with the constants K
r
and C
r
denoting the spring stiffness and damping in the
rotor shaft, and similarly; K
g
and C
g
as representing the spring stiffness and damping in the
generator shaft. Figure 2 also shows the inertia, torque, rot. speed and displacement of the
rotor and generator shafts. The parameters named as T1,
1
, q1, N1,I1 are the torque, speed,
displacement, number of teeth, and inertia of gear 1, and similarly for gear 2.
Fig. 2. Model of the drive train with the high and low speed shafts. (For denition of the
parameters, see text)
The model in Figure 2 results in the following equation of motion for the rotor torque
T
r
= I
r
q
r
+ K
r
(q
r
q
1
) + C
r
( q
r
q
1
) + T
1
+ I
1
q
1
(2)
where the factor K
r
(q
r
q
1
) + C
r
( q
r
q
1
) is the reaction torque in the low speed shaft.
Equivalently, the equation for generator motion is as follows
T
g
= I
g
q
g
+ K
g
(q
g
q
2
) + C
g
( q
g
q
2
) + T
2
+ I
2
q
2
(3)
where the factor K
g
(q
g
q
2
) + C
g
( q
g
q
2
) is the reaction torque at the high speed shaft.
The relationship between T
1
and T
2
is derived based on the equation describing a constrained
motion between two gears in the following way
T
2
T
1
=

1
2
T
1
=

1
2
T
2
(4)
From Eqn. (3) we nd
T
2
= T
g
I
g
q
g
K
r
(q
g
q
2
) C
r
( q
g
q
2
) I
2
q
2
(5)
then, the following equation for the rotor rotation holds
T
r
= I
r
q
r
+K
r
(q
r
q
1
) +C
r
( q
r
q
1
) +

1
2
(T
g
I
g
q
g
K
r
(q
g
q
2
) C
r
( q
g
q
2
) I
2
q
2
) + I
1
q
1
(6)
Since the goal of the modeling is to use it for control design, this equation will be simplied
in the following. First, it can be assumed that the the high speed shaft is stiff. This will imply
that T
g
= T
2
,
g
=
2
and so on. Secondly, the gearbox can be assumed lossless, hence the
terms involving I
1
and I
2
can be omitted. This reduces Eqn. 6 to the following equation
T
r
= I
r
q
r
+ K
r
(q
r
q
1
) + C
r
( q
r
q
1
) +

1
2
(T
g
I
g
q
g
) (7)
The model is nowof three degrees of freedom(DOF); rotor speed, generator speed, and a DOF
describing the torsional spring stiffness of the drive train. These DOFs correspond with the
three states shown in Figure 3
Fig. 3. Illustration of the 3-state model used in the control design having K
d
and C
d
as drive
train torsional stiffness and damping constants, respectively
The states in Figure 3 will in the following be regarded as perturbations from an steady-state
equilibrium point (operating point), around which the linearization is done. Hence the states
are assigned with the notation and describes the following DOFs
X
1
=
r
is the perturbed rotor speed
X
2
= K
d
(q
r
q
g
)is the perturbed drive train torsional spring stiffness
X
3
=
g
is the perturbed generator speed
where q
r
=
r
and q
g
=
g
.
Now, from the Newtons second law, the following relation holds
I
r
q
r
= T
r
T
sh
(8)
where the left hand side expresses the difference between the aerodynamic torque on the
rotor caused by the wind force, and the reaction torque in the shaft. This reaction torque can
be expressed according to Eqn. (7) as
T
sh
= K
d
(q
r
q
g
) + C
d
( q
r
q
g
) = K
d
(q
r
q
g
) + C
d
(X
1
X
3
) +

1
2
(T
g
I
g
q
g
) (9)
This equation can, when expressed in terms of deviations from the steady state operation
point, be written as:
T
sh
= K
d
(q
r
q
g
) + C
d
( q
r
q
g
) +

1
2
(T
g
I
g
q
g
) (10)
Hence the BEM theory provides a way to calculate the power coefcient C
P
based on
the combination of a momentum balance and a empirical study of how the lift and drag
coefcients depend on the the collective pitch angle, , and tip speed ratio, . In this way
an expression of the aerodynamic torque can be found to be
T
r
(V,
r
, ) =
1
2
R
2
C
P
(, )
V
2
(11)
where is the air density, R is the rotor radius, and V is the wind speed.
Let us now assume an operating point at (V
0
,
r,0
,
0
) such that Eqn. (11) can be written as
T
r
= T
r
(V
0
,
r,0
,
0
) + T
r
(12)
where T
r
is deviations in the torque from the equilibrium point (T
r
= T
r
T
r,0
) and consists
of partial derivatives of the torque with respect to the different variables, i.e Taylor series
expansion (Henriksen, 2007) and (Wright, 2004), in the following way
T
r
=
T
r
V
V +
T
r
r
+
T
r
(13)
where V = V V
0
, =
r

r,0
, and =
0
. By assigning , , and to denote
the partial derivatives of the torque at the chosen operating point (V
0
,
rot,0
,
0
), Eqn. (12)
becomes
T
r
= T
r
(V
0
,
r,0
,
0
) + (V) + (
r
) + () (14)
If the above expression is put into Eqn.(8), it follows that
I
r
q
r
= T
r
(V
0
,
r,0
,
0
) + T
r
T
sh,0
T
sh
(15)
At the operation point is T
r
(V
0
,
r,0
,
0
) = T
sh,0
since this is a steady state situation. This
reduces Eqn. (15) to
I
r
q
r
= (V) + (X
1
) + () X
2
C
d
(X
1
X
3
) (16)
when substituting with the corresponding state equations. The following expressions for the
derivatives of the state variables can now be set up
X
1
=
( C
d
)X
1
X
2
+ C
d
X
3
+ () + (
r
)
I
r
(17)
X
2
= K
d
( q
r
q
g
) = K
d
X
1
K
d
X
3
(18)
X
3
=
C
d
X
1
+ X
2
C
d
X
3
I
g
(19)
Note that in the derivative of the generator speed state X
3
is it used that I
g
q
g
= I
g
X
3
=
T
sh
T
g
= T
sh
when assuming a constant generator torque.
The dynamic system can now be represented in a state space system on the form as described
by Eqns (1) yielding
X
1
X
2
X
3
C
d
I
r

1
I
r
C
d
I
r
K
d
0 K
d
C
d
I
g
1
I
g

C
d
I
g
X
1
X
2
X
3
I
r
0
0
I
r
0
0
V
y =
0 0 1
X
1
X
2
X
3
(20)
where the disturbance input vector u
D
from the general form in Eqns. (1) now is given as V,
which is the perturbed wind disturbance (i.e deviations from the operating point, V = V
V
0
), and the control input vector u from Eqns. (1) now given as (i.e perturbed (collective)
pitch angle, =
0
).
The parameters will be assigned when coming to Section 4. It is worth no notice that the the
measured output signal here is the generator speed. This can be seen from the form of the
output vector C. It will be shown later that this will lead to a non-minimum phase plant, i.e a
plant with an asymptotically unstable zero in the right complex plane (Lee, 2004). Such plant
is in general not suitable for the LTR approach such that a revision of this is required.
3. Wind turbine control
3.1 Introduction
In order to have a power production which ensures that speed, torque and power are within
acceptable limits for the different wind speed regions, is it necessary to control the WT. This
control system should be complex enough to meet the intended control objectives, but at the
same time simple enough to easy interpret the results. A frequently used approach, which
will also be applied in this work, is to start with a simple model and a simple controller that
can be developed further by adding more degrees of freedom into the model.
Optimally speaking, if the control system shall be able to meet the requirement of reduced
energy cost, it must nd a good balance between (a) a long working life without failures and
(b) an efcient (optimal power output) and stable energy conversion. The rst point can be
regarded as the main focus in this work. This requirement can be further crystallized to the
following properties which the control system should possess
good closed loop performance in terms of stability, disturbance attenuation, and reference
tracking, at an acceptable level of control effort.
low dynamic order (because of hardware constraints)
good robustness
To meet the objective of long working life should the control system be designed to be
efciently mitigate loads in order to reduce the fatigue stresses (especially in shaft and blade
rotor) due to varying wind disturbances. Before going into the specic control design, an
historical overview of WT control with a special view toward the objective of load reduction,
will be presented in the following.
3.2 History of WT control
The history of WT control and research within this eld have emerged from the simplest form
of passive stall control to advanced controllers, like so-called smart rotor control. The latter is
well described in for instance (Wilson, Berg, Barone, Berg, Resor &Lobitz, 2009a) and (Lackner
& van Kuik, 2010). As these references portrays does this control scheme involve active
aerodynamic ow control by implementation of numerous sensors and actuators, bringing
along a high level of complexity. Although these advanced control methods have been
investigated for ten to fteen years, most commercial systems are still implemented using
multiple single-input-single-output (SISO) loops with classical PID controllers (Bossanyi,
2000). Actually, as reported in (Bossanyi, 2004), PID showed to give competitive results
compared to some of the new advanced techniques.
The traditional way of controlling a WT with multiple control objectives, such as speed control
for maximum power tracking and load mitigation by pitch control, is to design independent
control loops in a way illustrated in Figure 4(a).
Fig. 4. Illustration of difference between SISO and MIMO controllers
The PID controllers (i.e. the SISO controllers shown in Figure 4 (a)) are traditionally used for
the individual torque and pitch control and have shown to have a good effect when carefully
tuned and adjusted to its specic application. One disadvantage is, however, that the PID
control loops must be designed not to interfere with each other. If this happen to be the case
will the result often be a destabilized turbine. This problem can be solved in an efcient
manner within the modern and so-called advanced control techniques (Wright & Fingersh,
2008) using MIMO controllers (cf. Figure 4 (b)). In these more advanced control designs,
multiple control objectives is seen to be met with fewer control loops leading to stable closed
loop behavior (Wright & Stol, 2008).
With increasing turbine sizes, much research is done to nd new and better ways of load
control compared to the classical methods (see (Wright & Stol, 2008) and (Bossanyi, 2003) with
references). Large turbine sizes will give rise to loads that vary along the blade and change
quickly due to wind gusts and other varying wind conditions. Rapidly changing loads can
cause fatigue damage and reduce the life of the WT, which in turn may decide the lifetime of
the other turbine components. Because of the inertia of the system, as well as the limitations
of the actuators, active pitch control alone can only control Saverage
T loads on the blade.

On the other hand, passive load control strategies cannot respond to local load variations.
Active aerodynamic load control (AALC) is therefore suggested to have a good potential as
an addition the existing control strategies when it comes to load reduction (Wilson, Berg,
Barone, Berg, Resor & Lobitz, 2009b). One approach where AALC has been combined with
an individual blade pitch control scheme has shown to reduce the root ap bending moment
signicantly (Wilson, Berg, Resor, Barone & Berg, 2009).
A well known MIMO controller, the Linear Quadratic Gaussian (LQG) regulator, is suggested
in (Selvam, 2007) to be used in load control. This paper presents that this regulation policy
can have a good load reduction capability for a large frequency range. Reports as (Bossanyi,
2003) and (Bossanyi, 2004) suggest load mitigation when using individual pitch actuators. A
thorough study has been performed by Bossanyi in (Bossanyi, 2003) where the classical PI
control is compared with a multi-variable LQG control approach. Although this work has
shown to yield good results when applying LQG, the design process is not straightforward
and the resulting algorithm is relatively difcult.
One of the disadvantages of LQG design, however, is that it cannot directly take into account
robustness margins (like the gain- and the phase margin). Due to this poor robustness
properties of LQG control it was been necessary to search after other methods which could
handle model uncertainties in a better way. One method which can be traced back to the early
1980s is the H
approach. In this approach does the control designer from the very beginning
specify a model of system uncertainties, which can be for instance additive perturbations or
output disturbance (D.-W.Gu et al., 2005). Connor, Iyer, Leithead and Grimble were among
the rst to suggest an application of H
control on a WT model (see Connor et al., 1992). Their

main concern in this work were how to reduce the matter of fatigue loading. Although they
encountered some overshoot problems did they prove that the method was applicable to a
WT control problem. Suggestions of how H
control can be used for load reduction has later

been reported in (Bianchi et al., 2004). The method has also been shown to be applicable in
advanced power control to enhance a better power capture for a wider range of wind speeds
(Liu et al., 2008).
One particular problem the WT control designers must be able to handle is that WT control
generally involves a multi-objective optimization problem, each objective having different
goals. For instance, the control objective can be alleviate loads due to large scale gusts over
the whole disk area, while others can be more slowly varying and low frequency loads over
only one blade. To handle this problem does (Bottasso et al., 2010) propose an approach based
on a multi-layer architecture. In this method are three control layers designed, each aiming at
a specic control target and cooperates with the other layers to obtain various control goals.
In this way, the choice of the controller used on each layer can be optimized and tailored to
the specic control goal of that layer, thereby improving performance and simplify tuning
(Bottasso et al., 2010).
What the control designer in all of the above mentioned controlling methods are actually
trying to do is to compensate for the stochastic uctuations in the wind. To do this in a perfect
way is of course more or less impossible. One method called Disturbance Accommodating
Control (DAC) take into account the uctuations in the wind by an estimation of the
disturbance for an assumed above-rated wind speed situation (Balas et al., 1998). DAC
has shown to have a good effect on load mitigation and is therefore used to obtain special
attention when it comes to fatigue load reduction. The theory was thoroughly compared with
PI control by Alan Wright in his PhD thesis from 2004 (Wright, 2004). He concluded in his
paper that the analytic results obtained using DAC appeared to be "extremely promising". For
instance, it was shown how DAC has much better performance compared to the PI control
in reducing drive-train shaft torsional moments and blade root apwise-bending moments
(Wright, 2004). A brief description of the theory behind the DAC will be given shortly, while
an implementation and testing of DAC for the WT system will be done in Section 4.
In addition to the above mentioned control schemes are methods as Model Predictive Control
(MPC) (Henriksen, 2007), Generalized Predictive Controller (GPC), and Fuzzy Logic Control
(Karimi-Davijani et al., 2009) proposed for WT control. A comparison of different control
techniques will not be done in this work (please refer to (Bottasso et al., 2007) for more on
this).
3.3 The Linear Quadratic Regulator (LQR)
The LQR controller was among the rst of the so-called advanced control techniques used in
control of wind turbines. Liebst presented in 1985 a pitch control systemfor the KaMeWa wind
turbine using LQR design (Wright, 2004). The objective of this controller was to alleviate blade
loads due to wind shear, gravity, and tower deection using individual blade pitch control.
Results of this work included a reduction of blade and tower cyclic responses as well as the
reduction of a large 2-per-revolution variation in power (Liebst, 1985).
The practical application of this technique is limited by the challenges of obtaining accurate
measurements of the states needed in the controller. One way to solve this is to add
extra sensors and measurement arrangements. However, since this extra sensors and
measurements will add considerable cost and complexity to a WT, is this most often not
regarded as a good solution. In addition, errors in the measurement of these states can result in
poor controller behavior (Wright, 2004). To avoid these problems, a state estimator is added in
any practical LQRimplementation (Laks et al., 2009). This approach was described by Mattson
in his Ph.D thesis as early as in 1984, where power-regulation for a xed-speed WT using
blade pitch is presented (Mattson, 1984). The proposed solution to the WT control problem
was here based on linear models containing drive-train torsion and tower fore-aft bending
DOF. The same report also describes the use of state estimation to estimate wind speed.
The principles of a LQR controller is given in Figure 5. Here the state space system is
represented with its matrices A, B, and C, together with the LQR controller (shown with the
K).
The LQR problem rests upon the following three assumptions (Athans, 1981):
Fig. 5. State space control using a LQR controller where K is the LQR gain matrix
1. All the states are available for feedback, i.e it can be measured by sensors etc.
2. The system are stabilizable which means that all of its unstable modes are controllable
3. The system are detectable having all its unstable modes observable
LQRdesign is a part of what in the control area is called optimal control. This regulator provides
an optimal control law for a linear system with quadratic performance index yielding a cost
function on the form (Burns, 2001)
J =

0
x
T
(t)Qx(t) + u
T
(t)Ru(t) dt (21)
where Q = Q
T
and R = R
T
are weighting parameters that penalize the states and the control
effort, respectively. These matrices are therefore controller tuning parameters.
It is crucial that Qmust be chosen in accordance to the emphasize we want to give the response
of certain states, or in other words; how we will penalize the states. Likewise, the chosen
value(s) of R will penalize the control effort u. As an example, if Q is increased while keeping
R at the same value, the settling time will be reduced as the states approach zero at a faster
rate. This means that more importance is being placed on keeping the states small at the
expense of increased control effort. On the other side, if R is very large relative to Q, the
control energy is penalized very heavily. Hence, in an optimal control problem the control
system seeks to maximize the return from the system with minimum cost.
In a LQR design, because of the quadratic performance index of the cost function, the system
has a mathematical solution that yields an optimal control law given as
u(t) = Kx(t) (22)
where u is the control input and K is the state feedback gain matrix given as K = R
1
B
T
S. It
can be shown (see (Burns, 2001))that S can be found by solving the following algebraic Riccati
Equation (ARE)
SA + A
T
S + Q SBR
1
B
T
S = 0 (23)
The process of minimizing the cost function by solving this equation can easily be done using
the built-in MATLAB function lqr.
As can be seen from the above description does the LQR algorithm take care of the tedious
work of optimizing the controller, which for the case of the PID controller can be a time
consuming task. However, the control designer still needs to specify the weighting factors
and compare the results with the specied design goals. This means that controller synthesis
will often tend to be an iterative process where the designed "optimal" controllers are tested
through simulations and will be adjusted according to the specied design goals.
The optimal control solution from the LQR approach yields some impressive properties.
Firstly the controller guarantees at least 60
phase margin in each input channel (Saberi

et al., 1993), which means that as much as 60
phase changes can be tolerated in each input

channel without violating the stability. In addition, the LQR has innite gain margin which
implies that the LQR is still able to guarantee the stability, even though the gain increases
indenitely. Hence, LQR guarantees a very robust controller, capable of handling various
kinds of uncertainties in the modeling. However, when an state estimator is implemented
with measurement feedback, these properties just described is often lacking. But, as will be
shown later; the robustness properties can be recovered!
3.4 State estimation
As mentioned for the case of the LQR controller, all sensors for measuring the different states
are in this case assumed to be available. This is not a valid assumption in practice. A void of
sensors means that all states (full-order state observers), or some of the states (reduced order
observer), are not immediately available for use in the control task. In such cases, a state
observer must be implemented to supply accurate estimations of the states at all wind-rotor
positions. The observer state equations are given by
x = A x + Bu + L(y C x) (24)
y = C x (25)
where x is the estimate of the actual state x. Furthermore, equations (24) and (25) can be
re-written to become
x = (A LC) x + Bu + Ly (26)
These are the governing equations for a full order observer, having two inputs u and y and one
output, x. Since we already know A, B, and u, observers of this kind is simple in design and
provides accurate estimation of all the states around the linearized point. The proportional
observer gain matrix, L, can be found for instance by pole placement procedure, or as will be
shown in the following; by applying the Kalman lter design.
3.5 Kalman estimator
In the previous design of the state observer, the measurements y = Cx were assumed to be
noise free. This is not usually the case in practical life. Other unknown inputs yielding the
state equations to be on the general stochastic state space form, which is already dened in
Eqn. (1) as
x = Ax + Bu +u
D
y = Cx + Du + n
here the D matrix relating the input u to the output y most commonly is set to zero, the
disturbance input vector u
D
is stationary, zero mean Gaussian white process noise (i.e wind
disturbance), and n is zero mean Gaussian white sensor noise.
The Kalman lter can be applied to yield the estimated state vector x and output vector y (cf.
Eqns (24) and (25), by using the known inputs u and the measurements y. The Kalman lter
is optimal in the sense that it minimizes the estimated error covariance when some presumed
conditions are met (see Bottasso & Croce 2009). The mean-square estimation error is given by
J = E[(x(t) x(t))
T
(x(t) x(t))] (27)
where
E[(x(t) x(t))
T
y(t) = 0 (28)
The optimal Kalman gain is given by
L(t) = S
e
(t)C
T
R
1
(29)
where S
e
(t) is the same as J given in (3.5). Furthermore, when t , the algebraic Riccati
equation can be written as
0 = S
e
A
T
+ AS
e
+ Q
n
S
e
C
T
R
1
n
CS
e
(30)
where Q
n
and R
n
are the process and measurement noise covariances, respectively. Tuning of
the Kalman lter are required if these are not known. Finally, the sub-optimal Kalman gain
for a steady state Kalman lter can be expressed as L = S
e
C
T
R
1
.
3.6 Disturbance accommodating control
This section will explain the principles of DAC and for more details, please refer to e.g.
(Johnson, 1976). The DAC method will be utilized together with a LQR regulator and is a
straightforward way to model and simulate a system with an assumed-wind disturbance.
Recalling fromthe state space systemdescribed by Eqns. (1), which included the vector . This
vector describes the magnitude of the disturbance, while the input signal u
D
is the disturbance
quantity (which in our case is the wind speed perturbation). The vector will in the DAC
procedure be used as a free design parameter.
The basic thought in DAC theory is that disturbances are described by an assumed-waveform
model on state space form. Such a wind model can be described according to the following
equations (Wright, 2004)
z
D
= Fz
D
(31)
u
D
= z
D
(32)
where z
D
denotes the disturbance state, F the state matrix, u
D
is disturbance output vector
(input to the plant), and relates disturbance model input to disturbance states. Note that
it is necessary to know something about the initial condition of the disturbance state, z
0
D
, in
order to know the amplitude of the disturbance.
Since the state variables x and the disturbance variable z
D
cannot be measured directly,
observers will be utilized to estimate these. The mathematical description of the observer
will in this case according to Eqns (24) and (25) be given as
x = A x + Bu + u
D
+ K
x
(y y) (33)
y = C x (34)
where the estimator gains K
x
can be chosen according to for instance the pole placement
method using the place command in MATLAB.
Furthermore, the disturbance observer is described as
z
D
= Fz
D
+ K
D
(y y) (35)
u
D
= z
D
(36)
where the disturbance state estimator gain K
D
is a scalar.
The estimated variables can then be fed into the controller to minimize the effect of the
disturbances. The result of this is that the state feedback now also include the feedback of
disturbance observer yielding a "new" feedback control law on the form
u(t) = G x(t) + G
D
z
D
(t) (37)
where the LQR gain G is computed in MATLAB, and the full state feedback gain (i.e
disturbance gain) G
D
must be chosen as to minimize the norm |BG
D
+ |. Calculated
numerical values will be given in Section 4.
3.7 The LQG controller
Another well-known method of handling noise inputs is the Linear Quadratic Gaussian
(LQG) controller. This is simply a combination of a Kalman lter and a LQR controller.
The separation principle guarantees that these can be designed and computed independently
(Morimoto, 1991). LQG controllers can be used both in linear time-invariant (LTI) systems as
well as in linear time-variant systems. The application to linear time-variant systems enables
the design of linear feedback controllers for non-linear uncertain systems, which is the case
for the WT system.
The schematics of a LQG is in essence similar to that depicted in Figure ?? in Section 3.4,
the only difference is that the observer gain matrix L in this gure now can be dened as the
Kalman gain K
f
. However, we also assume disturbances in formof noise, such that the system
with the LQG regulator in compressed form can be described as in Figure 6
4. Control design and simulations
4.1 Introduction
The purpose in this project is to investigate the application of modern control theories, such
as the LQR and LQG, on WT systems and how a speed controller can be made with these
theories to reduce torque variations due to a wind disturbance. In addition, a further idea is
to investigate if a better load reduction can be obtained by using a dynamic compensator from
the LTR algorithm, and whether factors as the LTR gain will affect this result.
The simulations are done in the MATLAB Simulink environment, enabling the LTI state space
model described in Section 2 to be incorporated together with the different controllers in the
convenient block diagram approach available in Simulink.
Fig. 6. LQG regulator
This Section presents results from the practical implementation of the WT control design. It
is important to note that a test of the controlled system under various operating conditions
has not been performed. This is mainly due to the fact that this work is not focused on a
real-life application of the controller, but instead on the theoretical aspects of controller design.
Also, the linearization of the WT plant at different operation points would have required a
computer program, such as FAST
1
. Load mitigation in above rated wind speeds has already
been described to be the focus of this work, and the operation point chosen will reect this
goal.
The rst control approach that is being tested is the DAC theory, where the wind disturbance
modeled in a straightforward way together with the LQR and observer procedures, as
described in Section 3. The reason for starting with this method is to investigate if the results
are comparable with the results from the LQG method, and if so; what is the difference in the
results, and which results can be regarded as the "best" results.
It is important to note that for the simulations in this project are the actuator constraints, such
as the dynamics of the pitch-system, not taken into consideration.
4.2 Linearization
As was described in Chaper 2 does the WT compose of a complex non-linear relationship
between factors such as the aerodynamic properties of the blades, pitch angle, and wind
speeds. The best approach is thus to build the model out from experimental data and a
non-linear simulation model of the WT. Although it is possible to build a non-linear WT
simulation model also in Simulink and extract linear models by utilizing the built-in trim
and linmod functions, has this not been done in this work. Since the mathematical model
presented in this paper is simplied a lot would this approach probably result in a more
inaccurate model than with the use of the tailored WT simulation code provided by FAST,
1
A popular aeroelastic design code for horizontal axis wind turbines developed by National Renewable
Energy Laboratory, USA (Jonkman & Buhl, 2005)
where many aspects regarding the aerodynamics and structural dynamics are taken into
account. The linearized state space model, which will be used in the control design is based
on what is presented in (Wright, 2004) and (Balas et al., 1998).
The operation point, around which the linearization is done, is given as (Wright, 2004),
V
0
= 18m/s
rot,0
= 42rpm
0
= 12
The corresponding FAST linearization with reference to (Wright, 2004) around this point
yields the following state space system
X
1
X
2
X
3
0.145 3.108 10
6
0.02445
2.691 10
7
0 2.691 10
7
0.1229 1.56 10
5 0.1229
X
1
X
2
X
3
3.456
0
0
1
0
0
(V)
y =
0 0 1
X
1
X
2
X
3
(38)
where the factors C
d
/I
rot
and C
d
/I
gen
are calculated by using the known relation to the other
parameters.
Recalling from Section 2 that vector is given as

I
r
, where is describes the relationship
between torque variations and wind variations,
T
r
V
. This means that if is increased, it will
follow that the torque is more sensitive to wind variations, which is surely a negative effect
when the matter of fatigue is concerned. The torque is given as a non-linear function in Eqn
(11), which shows the dependency on the wind speed, rotor speed and blade pitch, and will
be xed at the chosen operating point. However, the vector is initially chosen to [100]
T
.
The transfer function (TF) of the open loop systemformed by the linearized state space system
in Eqns. (4.2) is
G(s) =
1.499 10
14
s
2
0.4247s 1450
s
3
+0.374s
2
+503.3s +59.44
(39)
extracted by using the MATLAB function ss2tf
One of the zeros of this TF is located at s = 2.83 10
13
leading to a non-minimum phase
system. Since the LTR procedure applied as an extension to the LQG later on is not feasible
for non-minimum phase systems (Saberi et al., 1993) will this vector be formed as to represent
rotor speed measurement instead, which changes the output to
y =
1 0 0
X
1
X
2
X
3
(40)
4.3 Disturbance accommodating control
The DAC procedure has already been described in Section 3.6. The Simulink block digram
shown in Figure 7 was set up according to the governing equations described in this section.
Fig. 7. Simulink block diagram of DAC
The assumed waveform model is shown in Figure 7. It is assumed a step input disturbance
fromthe wind, hence F = 0 and = 1 will be chosen to reect this (Henriksen, 2007). In order
to "excite" the controller an initial condition (chosen as 1) were inserted into the integrator of
the disturbance model.
4.3.1 Simulation without any control
It is interesting rst to have a look at the response without any control at all. Then,
after implementing the controller, is it easier to see how good the controller works in the
disturbance accommodating task.
Now, by setting both the LQR and the wind disturbance gain equal to zero, perturbations in
rotor speed (i.e. deviation from the operation point) shown in Figure 8 when simulating for
50 seconds. According to the simulation result, it is observed that the wind disturbance is not
accommodated, and the rotor speed will increase.
4.3.2 Simulation with only wind disturbance gain
The next scenario which is to be simulated is a situation with the LQR control gain set to 0, but
with the wind disturbance state gain G
D
set to minimize the norm |BG
D
+|. The response
of the rotor speed is shown in Figure 9 below.
0 10 20 30 40 50
0
1
2
3
4
5
6
7
8
Time [sec]
P
e
r
t
u
r
b
a
t
i
o
n
s

i
n

r
o
t
o
r

s
p
e
e
d

[
r
p
m
]
Fig. 8. Perturbations in rotor speed with no control
0 10 20 30 40 50
0
1
2
3
4
5
6
7
8
Time [sec.]
P
e
r
t
u
r
b
a
t
i
o
n
s

i
n

r
o
t
o
r

s
p
e
e
d

[
r
p
m
]
No control
With dist. gain, no LQR
Fig. 9. Perturbations in rotor speed with disturbance gain
As can be seen in this plot does G
D
alone manage to regulate the speed, but with a long settling
time.
4.3.3 Simulation of DAC with LQR
The nal DAC approach will involve a LQR design. Hence, this design will be described
in the following, with reference to the theories on LQR presented in Section 3.3. The initial
parameters of the wighting functions Q and R are chosen arbitrarily. The next task is then
to simulate to check whether the results correspond with the expected performance. After an
iterative study while changing Qand R values, the following weighting matrices were chosen:
Q =
100 0 0
0 0 0
0 0 100
(41)
The control weight of the performance index R was set to 1.
The chosen values in Q will result in a relatively large penalty of the states x
1
and x
3
. This
means that if x
1
or x
3
is large, the large values in Q will amplify the effect of x
1
and x
3
in the
optimization problem. Since the optimization problem are to minimize J, the optimal control
u must force the states x
1
and x
3
to be small (which make sense physically since x
1
and x
3
represent the of the perturbations in rotor and generator speed, respectively). On the other
hand, the small R relative to the max values in Q involves very low penalty on the control
effort u in the minimization of J, and the optimal control u can then be large. For this small R,
the gain K can then be large resulting in a faster response.
Since the systemhas been found to be observable the estimator poles can be arbitrarily placed.
Hence, the poles were chosen to be at -10, -30, and -50 yielding the estimator gain matrix
K =
17, 235
1.41 10
8
89.626
(42)
The disturbance state estimator gain K
D
is a scalar and chosen to 10.
It was found when doing these simulations and checking the results, that an exponential LQR
were more appropriate for the current application. The exponential LQR gain were found by
applying the code:
P=are(A+alpha*eye(3),inv(R)*B*B,Q);
Gdac=inv(R)*B*P;
where alpha is the exponential time-constant chosen to 4. This resulted in the gain matrix:
Gdac =
18.2087 1.2265 10
5
0.0425
(42)
Simulation of the time response of perturbations in rotor speed for ten seconds is shown in
Figure 10.
0 2 4 6 8 10
1
0
1
2
3
4
5
Time [sec.]
P
e
r
t
u
r
b
a
t
i
o
n
s

i
n

r
o
t
o
r

s
p
e
e
d

[
r
p
m
]
No control
DAC with LQR
Fig. 10. Comparison of perturbations in rotor speed when applying DAC with LQR
When looking at the plots in Figure 10, it is clear that the disturbance now is much better
attenuated while comparing with the rst two scenarios. The reason why the speed have this
undershoot is due to the right complex plane zeros the LQR controller gives to the closed loop
system.
It is also of interest to see how the control signal changes in the three previous situations. This
is shown in Figure 11.
As can be seen from Figure 11 does the pitch angle end up having a higher value. This is
an expected situation since it is now assumed a constant wind disturbance which must be
0 2 4 6 8 10
0.4
0.2
0
0.2
0.4
0.6
0.8
1
Time [sec]
P
e
r
t
u
r
b
a
t
i
o
n
s

i
n

p
i
t
c
h

a
n
g
l
e

[
d
e
g
]
No control
DAC Control
Fig. 11. Comparison of control signal for three previous scenarios
accounted for by an increase in pitch angle. Otherwise, the step in the wind would result in
a higher rotor speed, a situation we do not want to occur and the very reason for having a
control system in the rst place.
Another interesting signal is the error signal, which is the discrepancy between the "measured"
output signal y and the estimated signal y. This is shown in Figure 12 below.
0 0.5 1 1.5 2
0.005
0
0.005
0.01
0.015
0.02
0.025
Fig. 12. Error signal for a step input
The DAC simulations in this section have shown that this control method enables to mitigate
disturbance due to a wind assigned to affect the rotor state. However, the simulations show
that the settling time is relatively long (approximately 10 sec). Whether it is possible to
obtained a "better" result using a LQG controller, will be the task in the following section.
The disturbance will now be replaced by two step functions and the disturbance vector in
order to make the DAC more comparable with the LQG analysis. This gives the disturbance
input signal as shown in Figure 13.
The output response when applying this disturbance model is given in Figure 14
4.4 LQG controller
The control objective is the same as for the DAC; make the perturbations in the rotor speed as
small as possible in order to attenuate the wind disturbance. The wind disturbance is made
0 1 2 3 4 5
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Time
Fig. 13. Step signal for disturbance modeling
0 5 10 15 20
0.3
0.25
0.2
0.15
0.1
0.05
0
0.05
0.1
0.15
Time [sec]
P
e
r
t
u
r
b
a
t
i
o
n
s

i
n

r
o
t
o
r

s
p
e
e
d

[
r
p
m
]
Fig. 14. Output response when applying step input
by using the combination of two step functions as described earlier. This will give a step input
affecting the rotor state. Firstly, a situation without control is showed in Figure 15.
0 10 20 30 40 50
0
0.5
1
1.5
2
2.5
Time [sec]
P
e
r
t
u
r
b
a
t
i
o
n
s

i
n

r
o
t
o
r

s
p
e
e
d

[
r
p
m
]
Fig. 15. Response in rotor speed without any control
The sudden change in the rotor speed after one second is due to the step input function. The
gure show that without any control will the rotor speed increase to approximately 2 rpm
above the operational speed. After four seconds, when the disturbance is set to zero again,
will the speed then decrease and be brought to zero after approximately 50 seconds.
Calculation of the Kalman gain used in the LQGcontroller is according to the theory discussed
in Section 3. There are different methods to do this calculation in MATLAB. One is by applying
the kalman function, and another is the lqrc function. The latter is for instance used in
the LTR demo provided with the Robust Control Toolbox in MATLAB. When applying the
same weighting as for the LQR in the DAC approach in the lqrc function, and the weighting
functions
Q
n
=
30 0
0 30
R
n
= 1
representing the process and measurement noise covariances in the kalman function, did the
kalman function yield much better disturbance attenuation for a given control effort (shown
in Figure 16). The corresponding load attenuation for the two cases is shown in Figure 17
below
0 1 2 3 4 5 6 7 8 9 10
1.2
1
0.8
0.6
0.4
0.2
0
0.2
Time [sec]
P
e
r
t
u
r
b
a
t
i
o
n
s

i
n

p
i
t
c
h

a
n
g
l
e

[
d
e
g
]
Control signal with lqrc fn
Control signal with kalman fn
Fig. 16. Comparison of control effort while applying kalman and lqrc functions in
MATLAB
The differences in the output response can be seen through the Kalman gain matrix, which
when using the lqrc function were calculated to
Kf =
10.38
1.76 10
6
1.65
(40)
Similarly, the Kalman gain matrix when using the kalman function is
Kf2 =
57.60
3.87 10
6
1.07
(40)
0 2 4 6 8 10
0.02
0
0.02
0.04
0.06
0.08
0.1
0.12
Time [sec]
P
e
r
t
u
r
b
a
t
i
o
n
s

i
n

r
o
t
o
r

s
p
e
e
d

[
r
p
m
]
Disturbance attenuation with lqrc fn
Disturbance attenuation with kalman fn
Fig. 17. Comparison of rotor speed for similar control effort while applying kalman and
lqrc functions in MATLAB
The LQR gains were calculated with an exponential LQR with a time-constant set to 10, and
were given as
G =
25.78 2.68 10
5
5.70
(40)
The perturbation in rotor speed when comparing the LQG with the DAC response is shown
in Figure 18.
0 2 4 6 8 10
0.3
0.25
0.2
0.15
0.1
0.05
0
0.05
0.1
0.15
Time [sec]
P
e
r
t
u
r
b
a
t
i
o
n
s

i
n

r
o
t
o
r

s
p
e
e
d

[
r
p
m
]
LQG
DAC
Fig. 18. Comparison of perturbations in rotor speed for LQG and DAC
Figure 18 makes it clear that the effect of the two different control methods is quite different
when comparing the two output signals. The LQG controller will not force the disturbance
to zero in the same way as the DAC does, which must be regarded as a disadvantage of the
LQG. However, the DAC gives a much larger overshoot which is not very positive seen from
a controllers perspective.
Similarly, the perturbations in pitch angle for the two methods is shown in Figure 19.
This gure shows that the control effort is much higher for the DAC approach than for LQG.
To improve the robustness properties of the LQG will the LTR approach now be introduced
as a way to ensure good robustness in spite of uncertainties in the plant model.
0 1 2 3 4 5 6 7 8
1.2
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
Time [sec]
P
e
r
t
u
r
b
a
t
i
o
n
s

i
n

p
i
t
c
h

a
n
g
l
e

[
d
e
g
]
LQG
DAC
Fig. 19. Comparison of perturbations in pitch angle for the LQG and DAC
5. Conclusions and further works
The purpose of this thesis has been to reduce the loads on a WT for above rated wind
speeds (Region III) by speed control when applying different controlling methods. This
was rst performed by pitch regulation with DAC. This regulation policy showed to have
some regulation capacity, but also resulted in some bias in the control signal. It was found
when utilizing the DAC controller that although the wind disturbance was well mitigated,
the settling time was relatively long(approx. 10 sec). It could have been interesting to extend
the work also to include torque regulation in below rated wind speeds. This regulation policy
would aimat mitigation of speed and torque variations due to wind disturbances in Region II.
The LQG regulator showed to give good speed attenuation, but since DAC and LQG is quite
different approaches were a comparison between them not possible. It would have been of
major interest to extend the work to also considering pitch actuator constraints to see howthis
would have affected the results, especially the control input signal.
6. Acknowledgment
This work has been (partially) funded by Norwegian Centre for Offshore Wind Energy
(NORCOWE) under grant 193821/S60 from Research Council of Norway (RCN). NORCOWE
is a consortium with partners from industry and science, hosted by Christian Michelsen
Research.
7. References
Archer, C. L. & Jacobson, M. Z. (2005). Evaluation of global wind power, Journal of Geophysical
Research Vol. 110: p. 17.
Athans, M. (1981). The Linear Quadratic LQR problem, Massachusets Institute of Technology .
Balas, M., Lee, Y. & Kendall, L. (n.d.). Disturbance tracking control theory with application to
horizontal axis wind turbines, Proceeding of the 1998 ASME Wind Energy Symposium,
Reno, Nevada, 12-15 January pp. 9599.
Bianchi, F., Mantz, R. & Christiansen, C. (2004). Power regulation in pitch-controlled
variable-speed WECS above rated wind speed, Renewable Energy Vol. 29(No. 11): pp.
19111922.
Bossanyi, E. A. (2000). The design of closed loop controllers for wind turbines, Wind Energy
Vol.3: pp.149163.
Bossanyi, E. A. (2003). Individual blade pitch control for load reduction, Wind Energy Vol.
6: pp.119128.
Bossanyi, E. A. (2004). Developments in individual blade pitch control, EWEA conference-The
Science of Making Torque from Wind DUWIND Delft University of Technology April 19-21
2004 .
Bottasso, C. & Croce, A. (2009). Cascading kalman observers of structural exible and
wind states for wind turbine control, Technical report, Dipartimento di Ingegneria
Aerospaziale, Politecnico di Milano, Milano, Italy.
Bottasso, C., Croce, A., Devecchi, D. & Riboldi, C. (2010). Multi-layer control architecture for
the reduction of deterministic and non-deterministic loads on wind turbines.
Bottasso, C., Croce, A. & Savini, B. (2007). Performance comparison of control schemes for
variable-speed wind turbines, Journal of Physics: Conference Series 75, The Science of
Making Torque from Wind .
Burns, R. S. (2001). Advanced Control Engineering, Butterworth Heinemann.
Connor, B., Iyer, S., Leithead, W. & Grimble, M. (1992). Control of a horizontal axis
wind turbine using H-innity control, First IEEE Conference on Control Applications,
pp. 117122.
Gu D.-W., Petkov P.Hr. & Konstantinov, M. (2005). Robust Control Design With MATLAB,
Springer.
Ekelund, T. (1997). Modeling and Linear Quadratic Optimal Control of Wind Turbines, PhD thesis,
Chalmers University of Technology, Gothenburg, Sweden.
Elliot, A. S. & Wright, A. D. (2004). Adams/wt: An industry-specic interactive modeling
interface for wind turbine analysis, Wind Energy .
Henriksen, L. C. (2007). Model Predictive Control of a Wind Turbine, Informatics and
Mathematical Modelling - Technical University of Denmark .
Johnson, C. D. (1976). Theory of Disturbance- Accommodating Controllers, Advances in Control
and Dynamic Systems Vol. 12: pp. 387489.
Karimi-Davijani, H., Sheikholeslami, A., Livani, H. & Karimi-Davijani, M. (2009). Fuzzy logic
control of doubly fed induction generator wind turbine, World Applied Sciences Journal
Vol. 6 (4): pp. 499508.
Lackner, M. A. & van Kuik, G. A. M. (2010). The performance of wind turbine smart rotor
control approaches during extreme loads, Journal of Solar Energy Engineering Vol.
132(No.1).
Laks, J. H., Pao, L. Y. & Wright, A. D. (2009). Control of wind turbines: Past, present, and
future, University of Colorado at Boulder, USA .
Lee, C.-H. (2004). Stabilization of nonlinear nonminimum phase systems: Adaptive parallel
approach using recurrent fuzzy neural network, Vol. 34(No. 2): pp. 10751088.
Lee, J. & Kim, S. (2010). Wind power generations impact on peak time demand and on future
power mix, Green Energy and Technology 3: 108112.
Liebst, B. (1985). A pitch control system for the KaMeWa Wind Turbine, Journal of Dynamic
Systems and Control Vol. 107(No.1): pp.4652.
Liu, J.-H., Xu, D.-P. & Yang, X.-Y. (2008). Multi-objective power control of a variable speed
wind turbine based in h innite theory, International Conference on Machine Learning
and Cybernetics.
Jonkman & Buhl Jr M.L. (2005). Fasts user guide, National Renewable Energy Laboratory .
Morimoto, H. (1991). Adaptive LQG regulator via the Separation Principle, IEEE Transactions
on Automatic Control VOL. 35(No. I): 8588.
Saberi, A., Chen, B. M. & Sannuti, P. (1993). Loop Transfer Recovery, Analysis and Design,
Springer.
Mattson S.E. (1984). Modeling and Control of Large Horizontal Axis Wind Power Plants, PhD
thesis, Lund Institute of Technology, Lund, Sweden.
Selvam, K. (2007). Individual pitch control for large scale wind turbines.
Waltz, E. (2008). Offshore wind may power the future, Scientic American .
Wilson, D. G., Berg, D. E., Barone, M. F., Berg, J. C., Resor, B. R. & Lobitz, D. W. (2009a).
Active aerodynamic blade control design for load reduction on large wind turbines,
European Wind Energy Conference and Exhibition, 16-19 March.
Wilson, D. G., Berg, D. E., Barone, M. F., Berg, J. C., Resor, B. R. & Lobitz, D. W. (2009b). Active
aerodynamic blade control design for load reduction on large wind turbines, AWEA
Wind Power Conference .
Wilson, D. G., Berg, D. E., Resor, B. R., Barone, M. F. & Berg, J. C. (2009). Combined
individual pitch control and active aerodynamic load controller investigation for the
5mw upwind turbine, AWEA Wind Power Conference .
Wright, A. D. (2004). Modern control design for exible wind turbines, Technical report,
National Renewable Energy Laboratory.
Wright, A. D. & Fingersh, L. J. (2008). Advanced control design for wind turbines
part i: Control design, implementation, and initial tests, Technical report, National
Renewable Energy Laboratory.
Wright, A. & Stol, K. A. (2008). Designing and testing controls to mitigate dynamic loads
in the controls advances research turbine, Conference Paper 2008 ASME Wind Energy
Symposium .
12
Active Isolation and Damping of Vibrations for
High Precision Laser Cutting Machine
Andrea Tonoli
1,2
, Angelo Bonfitto
2
and Mario Silvagni
2
,
Lester D. Suarez
2
and Enrico Zenerino
2
1
Mechanics Department
2
Mechatronics Laboratory Politecnico di Torino
Italy
1. Introduction
High precision industrial machines suffer the presence of vibrations due to several noise
sources: ground vibration, acoustic noise, direct force disturbances. These are sources of
several problems at different levels and of different natures causing the performance losses
on sensitive systems (Crede, 1951), (Rivin, 1979).
In the last years the need of higher processing quality and throughput resulted in a
continuing demand for higher accuracy. Therefore active isolation and vibration damping
systems became mandatory to satisfy these requests (Preumont, 2002), (Hyde, 1997).
In general, machine supports are designed for high stiffness to obtain a robust machine
alignment with respect to its surroundings. However, in the presence of significant ground
vibration levels the support stiffness is commonly sacrificed to reduce the transmission to
the payload stage. Efforts to go towards these issues are recorded in several applications
and the solutions are different for any particular situation, depending on the nature of the
vibration sources, the amount of the disturbances and the machine environment.
Three main categories of possible approaches can be identified: passive, active and semi-
active configurations.
Completely passive solutions have almost reached their maximum potential which is still
not sufficient to satisfy stringent requirements. On the opposite, the exponential growth in
electronics and actuators fields made the use of active and semi-active isolation more
feasible. In particular, active control architectures allow to perform an effective isolation at
low frequencies, which is a common requirement for very demanding applications like
micrometer motion control, defect inspections, critical dimensions measurement and
overlay metrology.
Active control arrangements are provided with sensors, actuators and controllers
(Watters, 1988). Each of them can be classified depending on their technology and
physical working principle. For the application considered in this work, the main
categories of sensors used are: displacement, velocity, acceleration, strain and force. In the
same way, the most common types of actuators are: shape memory alloys, magnetic,
piezoelectric, magnetostrictive and magneto-rheological fluids actuators (Thayer, 1998).
The choice of sensors and actuators is strictly related to the type of application and
requirements and has also influence on the selection of the control strategies to be


244
applied. Typically the main control approaches are feedback, classical or model based,
and feed-forward technique, mostly with adaptive filtering of reference (Anderson, 1996).
Depending on the type of the controller, the system model can be used to support the
control design or can play itself a fundamental role on the control action (model based
strategies) (Beadle et al, 2002), (Sullivan, 1997).
This chapter is focused on the evaluation of an active isolation and vibration damping
device on the working cell of a micro-mechanical laser center, using active electromagnetic
actuators. To clarify the goal of this study it is important to point out that: a) the vibration
damping is defined as the reduction of the response amplitude of the system within a
limited bandwidth near the natural frequencies of the system; b) vibration isolation is
defined as the attenuation of the response of the system after its corner frequency to cut-off
all disturbances after that frequency allowing all signals below it to pass with no alterations.
The machine object of study is composed by two main parts: a frame support and a payload
stage where the laser cutting is performed. The system performance in terms of accuracy
and precision is reduced by the presence of two main vibration sources: the ground and the
stage itself. The active device should meet two main goals: the payload vibrations damping
and the reduction of the transmissibility of ground disturbances.
In this work the phases followed to design, realize and validate the device are illustrated
with a particular emphasis on the mechatronics aspects of the project.
A detailed analysis of the plant components is reported along with an exhaustive
explanation of the supports, actuation and sensing subsystems design criteria.
The actuation block consists in four electromagnetic Lorentz type actuators (two per axis) (Brusa,
2001). The absolute velocities of the frame support and of the stage are measured by means of
eight geophone sensors to determine the amount of the disturbances (Huan, 1985), (Riedesel,
1990). The considerations leading to the choice of this sensing system are reported along with the
related signal conditioning stage. The design of the supports between the ground and the frame
and of the connections between the frame and the stage is also explained.
Furthermore, all the subsystems described in the first part of the chapter are modeled along
with their interactions. The Lagrange equations approach is used to represent the system
behavior and in particular the links between the mechanical and electrical subsystems are
illustrated. The model includes the plant, the sensing, the control and the actuation blocks.
In particular, the mechanical subsystem is considered as a four degrees of freedom system.
Time and frequency domain computations are carried out from the model to evaluate
vibration levels and displacements and to identify which control parameters need to be
carefully designed to satisfy the requirements.
The last section gives details about the proposed control action and the validation of the
device. The control law consists in a couple of decentralized actions exerted along X and
Y -axis allowing to minimize the ground vibrations transmission and damp the payload
vibrations. A Lead-Lag control strategy, performed with a digital platform based on DSP
and FPGA, is used to compensate the high-pass band dynamic of the geophone sensors and
to damp the vibrations (Kuo, 1996), (Elliott, 2001). The payload isolation is achieved by
feeding the control block with the difference of frame and stage velocities and giving the
proper current command to the actuators.
The chapter concludes on the comparisons between simulation and experimental tests,
illustrating the validity of the model and the effectiveness of the proposed approach. In
particular, the performance of the vibration damping has been evaluated by using the
frequency responses between the actuators force and the payload velocities and the active

Active Isolation and Damping of Vibrations for High Precision Laser Cutting Machine

245
isolation by simulating numerically the disturbances coming from the ground and by
evaluating their transmission through all the system till the payload in closed loop
configuration.
2. System architecture
In this section of the chapter a full description of all the machine subsystems is provided.
The mechanical, electrical, electronic and control parts are identified and fully described
separately in the first part. Furthermore, since the project can be assumed as a classical
mechatronics application, the different blocks are analyzed in their interactions in order to
provide an overall view of the system.

Fig. 1. a) Picture of the machine. b) Sketch of the system. 1: Frame; 2: Stage (Payload); 3:
Actuators; 4: Frame Stage Springs; 5: Air springs; 6: Frame sensors; 7: Stage sensors.
Figure 1.a shows a picture of the laser cutting machine while in the sketch of Figure 1.b all the
components of the system are highlighted. The stage (payload) (2) consists in a granitic base
that can move freely within the work volume and is surrounded by four electromechanical
actuators (3) acting between the frame (1) and the stage. The machine is partially isolated from
the ground by means of four air springs (5). Four mechanical springs (4) are vertically placed
between the frame and the stage. The vibrations due to the machine process and coming from
the ground are measured on the payload and on the frame by means of eight velocity inertial
sensors (6, 7). A schematic representation of the actuators, sensors and springs position is
reported in Figure 2 where
GF
c and
GF
k represent the damping and the stiffness introduced
by the supports,
FS
c and
FS
k are the damping and the stiffness of the springs acting as
connections between the frame and the stage. Actuators and sensors are placed so that they
can be considered collocated in order to minimize the couplings between the axes actions by
keeping the proper alternation between resonances and anti-resonances in the system
dynamics. The main machine parameters and specifications are listed in Table 1.
The design phases have been performed considering the mechatronics nature of the system
and the interactions between the machine subsystems, illustrated in Figure 3. A classical


246

Fig. 2. XY plane view of the system. Frame-stage spring (
FS
k ,
FS
c ), electromagnetic actuator
(ACT), velocity sensor (Sens.), Frame-Ground spring (
FG
k ,
FG
c ).

Stage mass 1450 kg
Frame mass 300 kg
Maximum displacement of the stage 2.5 mm
Inertia of the stage along X -axis in YZ -plane 200 kg m
2
Inertia of the frame along X -axis in YZ -plane 100 kg m
2
Table 1. Main parameters and specifications of the machine.

Fig. 3. Block diagram of the system.


247
feedback behavior is performed: eight velocities are acquired by the sensors measurements
and elaborate with conditioning and filtering stages in order to feed the actuators with the
proper commands by means of power electronics action. The filtering stage consists in the
implementation of a Lead-Lag control strategy designed to fulfill the machine requirements
in terms of: a) active isolation from the disturbances coming from the ground and b)
damping of the vibrations generated by the machine processes.
2.1 Actuators subsystem
The actuation on the system is realized by means of four electromagnetic Lorentz type
actuators placed as illustrated in Figure 1 and Figure 2.
The actuator configuration is reported in Figure 4 (a) Picture, b) Sketch), A and B are two
permanent magnets while C indicates the coil.

Fig. 4. a) Picture of the actuator, b) Section view (A and B: permanent magnets, C:coil).
The force
ACT
F generated by each actuator is:

ACT
F BNli = (1)
where B is the magnetic field, N is the number of turns, i is the current flowing in the coil,
l is the coil length. The direction of the resulting force is illustrated in Figure 5. The amount
of required force for each actuator is equal to 200 N while the main parameters of the
designed actuator are reported in Table 2.

Coil thickness 6 mm
Coil length 3.3 mm
Coil active section 198 mm
2

Copper current density 12 A/mm
2
Coil length (l) 200 mm
Coil max actuation force (F
ACT
) 200 N
Number of turns (N) 263 -
Number of coils per axis 2 -
Table 2. Actuators main parameters.
a)
b)


248
The design of the actuators has been performed starting from the requirements of force and
maximum displacement of the stage, then a current density and the wire section have been
selected in order to perform a FEM analysis and to compute the magnetic field. Finally, once
known all the electrical parameters, the coil length l has been computed.

Fig. 5. Actuator force generation.
The actuators parameters have been identified experimentally. The resulting values are:
resistance 4.33 R = , 9.64 R mH = . The actuator transfer function can be expressed as:

1
1 1
( )
( )
ACT
L
G s
R
Z s sL R
s
L
= = =
+
+
(2)
The stationary gain ( 0) G s = is:

10
1
( 0) 20log 12.73 G s dB
R

= = =

(3)
The electrical pole
e
is:
449 72
e
R
rad s Hz
L
= = = (4)
The resulting actuator trans-conductance transfer function is reported in Figure 6.

Fig. 6. Actuator trans-conductance transfer function (magnitude and phase).


249
2.2 Springs and supports
The frame and the stage are connected in the vertical direction by means of four linear
springs placed as indicated by 4 in Figure 1 (
SF
c and
SF
k in Figure 2). The design has been
performed computing displacements and stresses with a FEM software, starting from the
following requirements:
infinite fatigue life;
stiffness
40 ;
40 ;
32500 ;
SFx
SFy
SFz
k N mm
k N mm
k N mm

damping
228 ;
228 ;
4313 ;
SFx
SFy
SFz
c Ns m
c Ns m
c Ns m

maximum displacement 2.5 ;
MAX
z mm =
The designed spring is made of harmonic steel and is characterized by:
length 125 ;
SPRING
l mm =
diameter 5 ;
SPRING
d mm =
maximum value of stress 500 .
MAX
MPa =
The supports chosen to provide the system with a partial level of isolation from the ground
are four air-springs (5 in Figure 1,
GF
k and
GF
c in Figure 2) consisting in resilient element air
and neoprene diaphragm. They are characterized by the following features:
Nominal natural frequency:
12.3 ;
12.3 ;
5.4 ;
GFx
GFy
GFz
f Hz
f Hz
f Hz

stiffness
450 ;
450 ;
500 ;
GFx
GFy
GFz
k N mm
k N mm
k N mm

damping
575 ;
575 ;
1700 ;
GFx
GFy
GFz
c Ns m
c Ns m
c Ns m

Transmissibility at resonance: 8:1;
The maximum load is equal to 545 kg;
The maximum air pressure is equal to 80 psi (5.5 bar).
2.3 Sensing subsystem
The disturbances on the plant are evaluated by measuring the velocities on the stage and on
the frame along X -axis and Y -axis by means of eight geophones placed as indicated in
Figure 2. They can be considered as the most common velocity inertial sensors to measure
seismic vibrations and can be classified as electromagnetic sensors that measure the velocity
and produce a voltage signal thanks to the motion of a coil in a magnetic field (Hauge et al,


250
2002). One configuration of the conventional geophones consists of a cylindrical magnet
coaxial with a cylindrical coil as shown in Figure 7. The coil is made up of a good conductor
like copper and is wound around a nonconductive cylinder to avoid the effect of the eddy
current that can be caused by the current induced in the coil. The wire diameter and the
dimensions of the holding cylinder are designed according to the application requirements.

Fig. 7. Geophone active configuration scheme. a) Coil and springs installation. b) Cross
section.
The internal core is a permanent magnet selected to give the highest possible magnetic field
density to maximize the induced voltage in the coil. The coil is attached to the casing of the
geophone by means of leaf springs (membranes); these springs are designed so as to
maintain alignment in the motion of the coil relative to the magnet keeping the lowest
stiffness possible in order to have a low resonant frequency for the geophone.
The reverse configuration (Figure 8) is realized using a coil fixed to the casing while the
moving mass is the permanent magnet. Since the mass of the magnet is heavier than that of
the coil, this configuration leads to a lower natural frequency, but the moving part is larger
and heavier.

Fig. 8. Geophone reverse configuration scheme.
Two geophone sensors are tested in the system: active sensor LF24 (configuration in Figure
8) and passive sensor SM6 (configuration in Figure 7). The LF-24 Low Frequency Geophone
is characterized by the following parameters: natural frequency at 1Hz, distortion
measurement frequency at 12Hz and sensitivity equal to 15V/(m/s).
The sensor chosen is the passive sensor SM6 because it allows to have an extreme low noise
though the output needs to be amplified by an active conditioning stage


251
The transfer function of the sensor response between the velocity of the casing and the
induced voltage in the coil can be written in the form

2
2 2
2
=
+ +
n n
Gs
TFG
s s
(5)
where
n
K m = is the natural frequency of the geophone, 2
n
C m = is the damping
ratio including the effect of the eddy current and G Bl = is the transduction constant, where
B is the magnetic field generated by the permanent magnet and l is the length of the coil.
Considering that the first natural frequency of the system is at about 1.8 Hz, close to the
geophone natural frequency, the sensor sensitivity cannot be simply modeled as a constant
value. Thus the transfer function of the geophone response must be identified to make the
result more reliable.

Fig. 9. Geophone damping ratio identification a) SM6 (Active configuration); b) LF24
(Reverse configuration).
SM6 is a passive velocity sensor with the following parameters: natural frequency is 4.5Hz,
sensitivity is 28V/(m/s). The damping coefficient of SM6 can be indentified as the same way
as LF24 in Figure 9.a and it is equal to 1.
Figure 9.b shows the LF24 transfer function expressed in Eq.5 identified experimentally.
When the damping ratio is equal to 1, the experimental results fit quite well the model.
Since the generated voltage is proportional to the rate of crossing of the magnetic field, the
output of the instrument will be proportional to the velocity of the vibrating body. A typical
instrument of this kind may have a natural frequency between 1 Hz to 5 Hz. The sensitivity of
such instrument may be in the range of 2 V/ms
1
to 3.5 V/ms
1
with the maximum
displacement limited to about 5 mm peak to peak (Thomson, 1981). When a geophone is used
to measure vibrations with a frequency below its natural frequency, the proof-mass tends to
follow the motion of the vibrating body rather than staying stationary. This motion of the
proof-mass following the casing reduces the relative motion which in turn reduces the induced
voltage. Here the sensitivity of the sensor (ratio between the voltage and the casing velocity)
becomes very small which restricts the performance of the sensor and limits the range of usage
of the instrument to frequencies above its corner frequency. It is important to mention that
b) a)


252
both displacement and acceleration can be obtained from the velocity sensor by means of the
integral and derivative computation provided in most of the signal conditioner units.
2.4 Electronics subsystem
In this section the subsystems related to sensor acquisition and conditioning, power
electronics and control implementation (Sensor conditioning, Power electronics, control
blocks in Figure 3) are illustrated.
The electronics system architecture is shown in Figure 10. The main characteristic of this
architecture is the serial communication input/output line that provides high noise
immunity, which can be useful when signals must travel through a noisy environment, such
as with remote sensors.

Fig. 10. Electronics subsystem.
The digital carrier is used like a buffer to provide the proper current level for the serial
communication. Here, multiples system buses manage data exchange between the main
serial communication core (FPGA) and the communication boards placed on the plant.
The communication boards are provided with one digital-to-analog converter (DAC) and
two analog-to-digital converters (ADC). The DAC is a 16-bit, high-speed, low-noise voltage-
output DAC with 30-MHz serial interface that is capable of generating output signal
frequencies up to 1 MHz. The ADC is a single channel 12-bit analog-to-digital converter
with a high-speed serial interface and sample rate range of 50 ksps to 200 ksps.
Control Unit (Control block in Figure 3)
The control module is supported by a DSP/FPGAbased digital control unit. Hence the
overall control implementation can be divided between the two digital devices in order to
fulfill different requirements: control strategy realization on DSP and serial communication
implementation on FPGA.
The overall control strategy is characterized with a nested and decentralized control
structure where only the outer loop is implemented on DSP while the inner loop (current
loop) is realized on the power module directly. In particular, the outer loop computes the


253
right reference for the inner one starting from required error compensation. The same
strategy is applied for each axis.
Sensors conditioning module (Sensor conditioning block in Figure 3)
The sensors conditioning module provides the output signal from geophone by means of an
instrumentation amplifier circuits. The component is configured for dual-channel operation,
in order to connect two geophones together. This conditioning circuit meets the needs of
high performance and portable instrumentation when small signal amplitude conditioning
brought out.
Figure 11, shows the circuit layout for dual-channel. R1A and R1B are the gain setting
resistors.
With the ADC input in the range of [0-3] volt and assuming the maximum magnitude of
noise on geophones measurement about 1000 m/s, the setting resistors are selected to
achieve a gain of 100.

Fig. 11. Instrumentation amplifier circuits AD8224. R1A and R1B are the gain setting
resistors.
Power electronics (Power electronics block in Figure 3)
The power module (Figure 12) is based on a trans-conductance amplifier instead of a
switching amplifier in order to avoid noise due to the switching frequency. This kind of
amplifier operates as a voltage-to-current converter whit a differential input voltage (voltage
controlled current source configuration).
Figure 12 shows the electronics layout that is divided in three main stages: a) the trans-
conductance amplifier (Trans Amp), b) the current amplifier (Curr Amp) and c) the
feedback resistor (Feedback).
The power module uses the voltages reference ( )
in
V from the control unit to generate the
proper current ( )
L
I to the load (electromagnetic actuator assumed as a RL load). The first
stage performs the current control by means of an operational amplifier that is unity-gain
stable with a bandwidth of 1.8MHz and it is internally protected against over-temperature
conditions and current overloads. The second stage is a classical current amplifier with
bipolar transistors in Darlington configuration to increase the current gain. The last stage
provides the feedback signal to ensure the desired current in the load. The power supply is
in the range of 30V.


254

Fig. 12. Power electronic based on a trans-conductance amplifier.
3. Modeling
The system has been modeled by using four degrees of freedom considering that the
dynamics along XZ and YZ planes are almost the same. As the vertical stiffness of the
flexural springs is very high, it can be assumed that there is no relative displacement
between stage and frame at vertical direction, which means the displacement in z axis about
stage and frame are the same. Both the stage and the frame are assumed as moving about
the frame mass center with the same rotating speed. Four flexural steel springs have been
used to link the stage to the frame, four air springs are placed at the bottom of the frame,
two actuators are working in series between the stage and the frame and two geophones are
used to measure the velocities of stage and frame respectively. The model reference frames
are defined in Figure 2 XY -plane view and in Figure 13, 16 YZ -plane view.

Fig. 13. YZ plane 4 dof kinematic relationships scheme.
The degrees of freedom of the model are:
[ ] ; ; ;
F F S
X y z q = (6)


255
that indicate the displacement of the frame alongY -axis andZ -axis, the rotation of the frame
(and stage) around the X -axis mass center and the stage displacement along its Y -axis.
Referring to the illustration in Figure 14, it is possible to obtain the formulation of the
velocity of a generic point S of the stage:

( )
0 0
cos sin
( )
S F S S S S
F F F F S S S S S S
V V q j FS k j
y j z k q z j y q k

= + + =

= + + + +
G G G G G
G G G G

(7)

Fig. 14. YZ plane 4 dof model scheme.
The kinetic energy T of the system can be expressed as:

2 2 2 2
1 1 1 1
2 2 2 2
S S S F F F
T m V J m V J = + + +
G G

(8)
where
S
m ,
F
m ,
S
J and
F
J are the mass and the rotating inertia of the payload (the stage S)
and of the frame F .
The potential energy U is obtained starting from the diagram reported in Figure 14:

( ) ( )
2 2
2
1 2
2 2
( ) ...
( ) 2
= + + + + +
+ + +
GFz F G GFy F G GFz F G
GFy F G SFy S
U k z d z k y h y k z d z
k y h y k q

(9)
Owing to the Rayleigh dissipation formulation, the damping of the system is:

( ) ( )
2 2
2
1 2
2 2
2
( ) ...
( ) 2
= + + + + +
+ + +

GFz F G GFy F G GFz F G
GFy F G SFy S
c z d z c y h y c z d z
c y h y c q

(10)
where each damping term
i
c is obtained starting from the experimental identification of
damping ratios
i
:


256
2
i i i i
c k m = (11)
The inputs of the system are: the force of the electromagnet actuators
act
F , the force of the
stage
S
F and the velocities from the ground in y direction
Gy
v and z direction
Gz
v v. The
output are the velocities
F
v of the frame and
S
v of the stage measured with geophones
sensors. Inputs and outputs are graphically represented in Figure 15.

Fig. 15. YZ plane 4dof model scheme input and output.
Using the Lagrange formulation is possible to write the equations of motion in the form:
{ } q Cq q F + + = (12)
where

( )
T
F F S
q y z q = (13)
is the vector of the generalized coordinates,

( )
T
Gy Gz S act
F v v F F = (14)
is the vector of the generalized forces and M is the mass matrix

0
0
0 0 0
0
0
0 0
0
tot S S S
tot S S
S S S S tot S S
S S S S
m m z m
m m y
M
m z m y J m z
m m z m

=

(15)


257
with
tot S F
m m m = + ,
2 2
0 0
( )
tot S F S S S
J J J m y z = + + + .
0S
y ,
0S
z are the initial position of the
stage. The matrix is symmetric and not diagonal because it takes into account the coupling
between the stage and the frame dynamics.
The stiffness matrix K is:

1 2
2 2 2
1 2 1 2
4 0 4 0
0 4 2 2 0
4 2 2 2 4 2 0
0 0 0 4
GFy GFy
GFz GFz GFz
GFy GFz GFz GFz GFy GFz
SFy
k k h
k k d k d
K
k h k d k d k d k h k d
k

+

=

+ + +

(16)
The damping matrix C is:

1 2
2 2 2
1 2 1 2
4 0 4 0
0 4 2 2 0
4 2 2 2 4 2 0
0 0 0 4

+

=

+ + +

GFy GFy
GFz GFz GFz
GFy GFz GFz GFz GFy GFz
SFy
c c h
c c d c d
C
c h c d c d c d c h c d
c
(17)
The selection matrix T of the generalized forces is:

1 2 1 2
4 0 4 0
1 0
0 4 0 4 0 0
4 2 2 4 2 2 0 0
1 1
0 0 0 0

=

GFy GFy
GFz GFz
GFy GFz GFz GFy GFz GFz
k c
k c
T
k h k d k d c h c d c d
(18)
In the state space formulation the equations of motion of the system can rewritten as:

.
X AX BU = + (19)
where the state vector X and the input vector U are:
{ } { }
,
T
T
G G Gy Gz S act
X q q y z U v v F F = = (20)
with A the state matrix, B the input matrix

1 1 1 1
0 0 0
,
0 0 0

= =

I
A M K M C M T B M T
I
(21)
The relationship between input and output can be represented as:
Y CX DU = + (22)
where Y is the output vector, C the output matrix and D the feedthrough matrix


258
{ } [ ]
0 0 0 0 1 0 1 0 0
, , 0
0 0 0 0 1 0 1 0 0

= = =

geo T
S F
geo
z
Y v v C D
z
(23)
4. Control design & results
The control action is designed to achieve two main goals: active isolation of the payload
from the ground disturbances and vibration damping during the machine work processes.
These two actions allow to operate on the stage without external disturbances. The XZ and
YZ are considered the same so the control laws along the two planes are equivalent.
Furthermore, from the control point of view, the adopted model is oversized respect to the
control requirements if the goal is the isolation of the payload. As a matter of fact, in this
case a two degrees of freedom model is sufficient while if also the dynamics of the frame is
supposed to be controlled, then a 4 dof model is necessary.
The considered system can be considered as intrinsically stable due to the presence of
mechanical stiffness between the payload and the frame allowing to obtain a negative real
part for all the eigenvalues of the system.
Root loci of the system in open and closed loop configurations are reported in Figure 16.

Fig. 16. Root loci of open loop (a) and closed loop (b) configurations (Circles: zeros; Crosses:
poles).
Since the system along YZ ( XZ ) presents one actuation point and a couple of sensors
(frame and stage velocities), a solution with a SISO control strategy is not feasible. A
simplest solution to this problem considers the difference between the measured velocities
as the feedback signal, so the system can be assumed as SISO and the control design
becomes simpler.
Figure 17 shows that the system dynamics has a peak at 1.8 Hz related to the stage and
higher modes related to the interaction of the stage with the frame and the ground at 10 Hz
and beyond.
The control action is focused on damping the mode related to the stage by adding on the
loop a lead-lag compensator.


259

Fig. 17. Vibration damping action. Transfer function from the actuator force to the difference
of frame and stage velocities ( ( )
S F ACT
q q F ). Open-loop vs Closed-loop. Solid line:
experimental; Dashed line: numerical.
The two actions can be expressed as:

LAG
LAG
LAG
LEAD
LEAD
LEAD
s z
C
s p
s z
C
s p
+
=
+
+
=
+
(12)
The
LAG
C action is used to improve the transient response at low frequency, while the
LEAD
C is useful to increase the stability margin of the closed-loop system.
Therefore the resulting Lag-Lead action allows to compensate the geophone the critical
phase behavior of the geophones and furthermore guarantees a quick damping action with
good levels of stability margins.
The experimental tests have been performed to validate the two control actions. Figure 17
shows the frequency response function in open loop and closed loop, numerical and
experimental calculated between the force of the actuator and velocity measured with the
geophone on the stage. The force acts both on the stage and the frame, the dynamics of both
the subsystems are visible. The vibration damping of the control action is validated on the
stage mode (1.8 Hz peak) and the good correspondence between the simulated and
experimental response is useful to validate the modeling approach.
A further demonstration of the correctness of the damping action is the velocity time
response reported in Figure 18. In this case the system is excited with an impulse from the
actuator and the velocity is measured on the stage. Numerical and experimental responses
are superimposed to provide a further validation of the model (the position time response is
not reported since the machine is not provided with displacement sensors and hence this
validation could not be possible to perform). Figure 18.a shows open loop response, Figure
18.b shows closed loop response while in Figure 18.c the force exerted by the actuators is
reported.


260

Fig. 18. Impulse time response, force from the actuator and velocity measured on the stage.
Open-loop (a), Closed-loop (b), Force exerted by the actuators. Solid line: experimental
results. Dashed line: numerical results.

Fig. 19. Vibration damping action. Transfer function from a force applied on the stage to the
velocity measured on the stage ( )
S STAGE
q F . Numerical response. Solid line: closed-loop;
Dashed line: Open-loop.
The excitation coming from the laser-axis action on the payload is controlled in an effective
way as shown in Figure 19 where the numerical transfer function between a force impulse
on the stage and the related measured velocity is reported.
The active isolation action is verified by simulating the excitation coming from the ground.
The experimental test in this case has not been performed since in reality it is difficult to
excite the machine from the ground in a controlled and effective way. Nevertheless the


261
model is reliable as proved in Figure 15 and the obtained results can be assumed as a good
validation of the control action.
Figure 20 illustrates that the closed loop system is capable to reject the disturbances coming
from the ground in an effective way.

Fig. 20. Active isolation action. Transfer function from a simulated ground velocity to the
velocity measured on the stage ( )
S G
q q . Numerical response. Solid line: Open loop
configuration. Dashed line: Closed loop configuration.
5. Conclusions
In this chapter the realization of an active isolation and damping of vibrations for high
precision laser cutting machine has been illustrated.
The literature analysis confirms that in the last decades active configurations had an
exponential growth due to their high performance level.
The proposed solution uses Lorentz type electromagnetic actuators since they can be easily
integrated in the mechanical structure and allow to reach good results.
A mechatronic approach has been used to manage the criticalities introduced by the
subsystems interactions since it provides an overall view of the system.
A 4 degrees of freedom model has been carried out by means of Lagrange equations
approach and was used to choose and set the control strategy. The proposed control
technique is a lead-lag compensator able to control the stage dynamics and isolate it from
the external disturbances.
The effectiveness of the vibration damping and the active isolation actions have been
validated experimentally by analyzing the plant behavior in open-loop and closed-loop
configurations with frequency and time responses. The results have been compared to the
numerical computations to prove the correctness of the modeling approach.
Future works will be focused on the tests of new control strategies and on the evaluation of
the adopted solutions compared to existing methods.


262
6. References
Anderson E., Leo D.J., Holcomn M.D, (1996). UltraQuiet platform for active vibration
isolation, Proceedings Smart Structures and Integrated Systems 2717, San Diego,
CA, pp. 436451.
Beadle B.M., Hurlebaus S., Gaul L., Stbener U. (2005). Active control strategies for vibration
isolation, in Proc. IUTAM Symp. Vibration Control of Nonlinear Mechanisms and
Structures 2005, Munich, Germany, pp. 91-100.
Brusa E., Carabelli S., Genta G., Maddaleno F., Silvagni M., Tonoli A. (2001). Voice coil
actuator for Active Vibration Isolation in Microgravity, 6th ISMST, Torino.
Crede, C. (1951). Vibration and shock isolation, John Wiley and Sons Inc., New York, USA
Elliott S.J. (2001). Signal processing for active control, Academic Press.
Hauge, G.S. Campbell, M.E.. (2004). Sensors and control of a space-based six-axis vibration
isolation system. Journal of Sound and Vibration 269. Pp 913931.
Huan, S. L., & Pater, A. R. (1985). Analysis and prediction of geophone performance
parameters. Geophysics, 50(8), 1221:1228.
Hyde T.T, (1997) An experimental study of active isolation. Proceedings of the 38th
AIAA/ASME/ASCE/AHS/ASC, Structures, Structural Dynamics and Materials
Conference, Orlando, FL, pp. 17631773.
Kuo S.M., Morgan D.R., (1996). Active noise control systems: algorithms and DSP
implementations, John Wiley & Sons.
Preumont, A. (2002). Vibration control of Active Structures. Kluwer Academic Publishers,
2
nd
Edition, Netherlands
Riedesel, M., Moore, R. D., & Orcutt, J. A. 1990. Limits of Sensitivity of Seismometers with
Velocity Transducers and Electronic Amplifiers. Bulletin of the Seismological
Society of America, 80(6), 1725:1752.
Rivin, E. I. (1979). Principles and criteria of vibration isolation of machinery. Trans. of the
ASME, Journal of Mechanical Engineering, 101:682692.
Sullivan J., Rahman Z., Cobb R., Spanos J. (1997). Closed-loop performance of a vibration
isolation and suppression system, Proceedings of the American Control
Conference, Vol. 6, Albuquerque, NM, pp. 39743978.
Thayer, D., Vagners, J., von Flotow, A., Hardham, C. and Scribner, K., (1998). Six Axis
Vibration Isolation Using Soft Actuators and Multiple Sensors, Proceedings of the
21st Annual AAS Guidance and Control Conference, Feb 1998.
Thomson, W. (1971). Theory of vibration with applications. George Allen and Unwin.
Watters B.G., Coleman R.B., Duckworth G.L., Berkman E.F. (1988). A perspective on active
machinery isolation, Proceedings of the 27th Conference on Design and Control,
Vol. 3, Austin, TX, pp. 20332038.
13
Bearings Fault Detection Using Inference Tools
Miguel Delgado Prieto, Jordi Cusid i Roura and
Jose Luis Romeral Martnez
MCIA Group, Technical University of Catalonia
Spain
1. Introduction
Electric motors are nowadays widely used in all kind of industrial applications due to their
robustness and ease of control through inverters. Therefore, any effort, with the aim of
improving condition monitoring techniques applied to them, will result in a reduction of
overall production costs by means of productive lines stoppage reduction, and increment of
the industrial efficiency. In this context, the most used electric machine in the industry is the
Induction Motor (IM), due to its simplicity and reduced cost. The analysis of the origin of
IMs failures exhibits that the bearings are the major source of fault (Singh et al., 2003), and
even a common cause of degradation in other kinds of motors as Permanent Magnet
Synchronous Machines. An IM failures percentage distribution, according to previous
studies (ODonell, 1986), is shown in figure 1, in order to highlight the bearings monitoring
importance.

Fig. 1. IM failures percentage distribution.
Focusing in bearings defects, these have been typically categorized as distributed or local.
Distributed defects include mainly surface roughness, waviness and misaligned races.
Localized defects, however, include cracks, pits and spalls basically.
A great deal of studies concentrate their efforts in localized defects detection, because these
defects are generally related with concrete fault indicators in the acquired signals, meanly

characteristics faults harmonics in machine vibrations. However, the issue related with the
capability to detect both, localized and generalized bearings faults, is an important issue that
should be attended (Cusido et al., 2009). Generalized degradation is difficult to detect
because the lack of consistent characteristic fault harmonics in the spectral decomposition of
any acquired signal. Therefore, it is difficult to inspect deeply motor bearings by analysing
only some specific fault harmonics due to fault signal complexity (Obaid et al., 2003), and in
the case of asynchronous motors the slip factor introduces additional difficulty. This fact
makes necessary the diagnosis support of additional signal analysis (Frosini et al., 2008). The
current trends in condition monitoring are related with the fusion of different features,
which provide the possibility to merge fault indicators from different physical magnitudes
(Cusido et al. 2010). This data fusion improves the diagnostic reliability, because fault
indicators that are not descriptive enough themselves can contribute to detect faults in
relation with others, especially if the features are extracted from different physical
magnitudes, which enhance the monitoring capabilities. However, it is necessary to find the
best and most useful fault indicators, and merge them by some kind of inference tool.
In this chapter, additional bearings fault features are introduced to be merged with
traditional vibrations fault indicators. Hence, features from machine vibration, stator
currents, stator common mode currents and acoustic emissions, are presented and related
with bearing faults. With all these fault indicators diagnosis different systems based on
multidimensional features fusion can be implemented. The bearing diagnosis capability and
reliability are easily increased making possible the bearing fault detection even if the fault is
localized or generalized. Regarding the inference tools for features fusion, it can be chosen a
wide variety of methods such as statistical rules, expert systems or artificial intelligent
techniques among others. In this chapter, two different inference tools are presented in
order to generate two different diagnosis systems: Look-up tables, representing one of the
easiest ways to merge information, and Fuzzy logic as a next step towards advanced
diagnosis systems based on artificial intelligence.
2. Basic theory
A brief introduction of each proposed physical magnitude for bearings fault detection is
presented next, with special attention to the specific bearings fault indicators.
2.1 Vibration analysis
Vibration analysis is one of the most extended condition monitoring techniques. Despite
being a reliable, well studied robust technique, it requires that the motor under test has a
vibration transducer installed. The measurements should be taken on the bearings, bearing
support housing, or other structural parts that significantly respond to the dynamic forces
and characterize the overall vibration of the machine. Therefore, the major disadvantage of
vibration monitoring is that requires access to the machine, and specific accelerometers
housing over the machine is sometimes required. For accurate measurements, sensors
should be mounted tightly on the machine, and expertise is required in the mounting,
condition that makes its online application expensive. Sometimes, other techniques without
this kind of restriction are preferred or required.
The single-point bearing defects imply certain characteristic fault frequencies which will
appear in the machine vibrations. The frequencies are predictable and depend on which
surface of the bearings contains the fault. There is one characteristic fault frequency

Bearings Fault Detection Using Inference Tools 265
associated with each of the four parts of the bearing. Vibration frequency components
related to each of the four basic fault frequencies; (1) Fundamental train frequency, (2) Ball-
spin frequency, (3) Ball pass outer race and (4) Ball pass inner race, can be calculated using
the following expressions (Bellini et al., 2008):
FIF =
1
2
[1
P
d
B
d
(1)
BSF =
1
2
P
d
B
d
_1 [
B
d
P
d
cos
2
_ (2)
BPF0 =
n
2
[1
B
d
P
d
cos (3)
BPFI =
n
2
[1 +
B
d
P
d
cos (4)
with:
n: Number of bearing balls
fr: Rotor speed
Bd: Ball diameter
Pd: Bearing Pitch diameter
: Contact angle of the ball on the race

Fig. 2. Main bearing design parameters, B
d
: ball diameter, P
d
: pitch diameter, : contact angle.
Regarding the roughness bearings defects, there is a wide variety of causes from
contamination of the lubricant to the shaft currents or misalignment. The generalized
roughness faults produce unpredictable broadband effects in the machines vibration
spectrum, but it seems to be feasible the detection by means of the temporal vibration signal
Root Mean Square (RMS) analysis. As some works and standards (Riley et al., 1999; Cabanas
et al., 1996) set out, a RMS vibration value evaluation of the motor also provides a good
indicator for motor health, allowing machine overall fault diagnosis.
2.2 Stator currents
A Motor Current Signature Analysis (MCSA) represents by the stator currents acquisition an
interesting alternative method with its own particularities and benefits (Cusido et al., 2007a);
the most interesting of them is to avoid accessing inside the motor making it easy to perform

its online fault analysis (Cusido et al. 2007b). It has been demonstrated (Schoen et al., 1995) that
the characteristic bearing fault frequencies in vibration can be reflected on stator currents. As a
result of motor airgap length variations due to bearings defect, flux density is influenced and
then an additional magnetic flux appears. This magnetic flux, and its variations associated to
rotor turning, creates additional components that can be found in the stator currents spectra
(Cusido et al., 2005). Using this method it has been widely demonstrated in the literature (El
Hachemi Benbouzid, 2000) that different faults like eccentricity, rotor asymmetry, stator
winding failures, broken bars and bearings damage can be diagnosed. The relationship
between the vibration frequencies and the current frequencies for bearing faults can be
described by equation (5). Therefore, by means of (5), it is possible to analyze the specific fault
harmonics in order to find abnormalities in their amplitude values.

bg
= |
c
_m
| (5)
with:
f
bg
: Electrical fault characteristic frequency
m: Integer
f
e
: Electrical supply frequency
f
v
: Vibration fault characteristic frequency {(1), (2), (3) or (4)}
It is well established that for bearing single-point defects, the characteristic stator current
fault frequencies are good fault indicators. Even so, it was discovered in several studies, that
for many in situ generated bearing faults, those characteristics fault frequencies are not
observable and may not exist at all in stator current (Stack et al., 2004.). But it is
demonstrated also that these same bearings faults have an effect over the motor eccentricity
(Basak et al., 2006), and these characteristics stator current faults frequencies are easily
detectable as sidebands over the fundamental motor current frequency. Therefore, the
evaluation of the bearings characteristics stator current faults frequencies is useful for
diagnosis proposes, because it can diagnose directly the bearing fault. But as a second
diagnosis step, the analysis of stator current fundamental sidebands, in order to detect
eccentricity, can be useful also for bearing diagnosis. However, it is necessary other fault
indicators in order to classify correctly between eccentricity fault caused by bearing fault or
eccentricity fault caused by other faults in the motor.
Regarding generalized bearing defects, previous works have shown the existing correlation
between vibration and currents RMS values (Riley et al., 1999). Although it is a complex
function that relates both magnitudes, this work tries to check the RMS currents reliability in
order to perform the motor status diagnose.
2.3 High frequency common-mode pulses
One of the biggest culprits for bearings failure are common-mode circulating currents (CMC).
The CMC are generated due to the inverter used to manage motors, because the inverter
creates common mode voltage as figure 3 shows. Each high dv/dt over the inverter
modulation implies a proportional current, which is propagated over the motor trough
different paths to the ground in order to turn back to the inverter (Muetze and Binder, 2007a).
The CMC travels around the motor (and load if it is not electrically isolated), due to the
capacitive effect that two conductive materials separated by means of some isolating
material (dielectric) can create. For instance, the capacitive effect produced between the coil
group and the chassis separated with air gaps in an induction motor.


Fig. 3. Common mode voltage generated with PWM modulation.
The capacitances created inside the motor have a very low value, so the motor intrinsically
gets filter the low frequency currents, but the high frequency currents see low impedance
paths (Binder and Muetze, 2008.). Some current travel over the shaft, that in an electrical
sense, find the bearing rail, lubricant and bearing ball capacitive coupling. The high
frequency CMC pulses current that contain an important amplitude value, provoke a
discharge over the capacitive coupling. This phenomenon is called EDM (Electric Discharge
Machining) (Kar and Mohanty, 2008). The CMC influences on the bearings degradation due
to the effect that every CMC discharge provoke over the lubricant that recover the bearing,
because the continually application of these discharges implies lubricant degradation. This
effect increases the contact between the bearings with the rail accelerating the final bearings
degradation.
As it is shown in figure 4a, circulating currents could follow different paths to the ground
through the stator windings or rotor. One important path of the circulating currents is
through the bearings (Muetze and Binder, 2007b). The electrical scheme of parasitic
capacitive couplings is shown also in figure 4b. This scheme represents the CMC path from
inverter to bearings. As it has been explained previously, the inverter generates common
mode voltage (V
mc
) and at the same time, generates common mode current (I
mc
) which is
propagated trough the wire (L
C
), motor (L
m
) and through the coupling effect between the
motor and chassis, and between the motor and rotor, this last ones cross finally the coupling
effect between the shaft and the bearings.
A temporal CMC acquisition and a single common-mode discharge are shown in figure 5.
These currents typically show a frequency range of mega-hertz with a period of micro-
seconds between bursts. CMC discharges provoke bearings lubricant degradation. This
effect provokes the contact between the bearings with the rail. Therefore, CMC discharges
amplitude is directly depending of the parasitic capacitances which are depending of the
lubricant state and the distance between bearings and rail mainly. Therefore, seems to be
possible the bearings diagnosis by means of the number of CMC pulses that surpassed a
prefixed amplitude threshold during a fixed time, in order to distinguish between fault and
healthy bearings (Delgado et al., 2009). Analyzing the number of CMC pulses that surpassed
a current amplitude threshold value, it is possible to see that a minor number of CMC pulses
surpassing the threshold, is significant of a degradation state of the bearings, because the
capacitive effect rail-lubricant-bearing needs a minor energy differential to allow an EDM.


a)

b)
Fig. 4. a) Main CMC paths over inverter-motor-load system. b) Electrical Scheme for
capacitive and parasitic couplings.
Therefore, the methodology consists in a first time acquisition over the stator CMC in a test
bench with healthy bearings. The amplitude of the CMC pulses decrease at the same time
that bearings degradation increase, so is necessary to specify a CMC pulses amplitude
threshold and count the number of pulses that surpasses this threshold during a fixed time.
Obviously, the time acquisition and the threshold value make depends the number of CMC
pulses counted. An acquisition time of tens of milliseconds, and a threshold over the 75% of
the maximum CMC pulses amplitude over healthy bearing, is enough to distinguish
between healthy and degraded bearings.
In this work, to limit the CMC acquired signal to only pulses flowing through bearings (the
responsible of balls degradation), a motor modification was introduced. All the ball bearing
under test were isolated from the motor stator frame but in a point connected to ground
through a cable where the pulses were measured. Bearings insulation was achieved by
surrounding the piece with a polytetrafluoroethylene (PTFE) flat ring with a hole
mechanized in it to let the cable pass through.
2.4 Acoustic Emissions
The Acoustic Emission Technique is a very promising tool that has practical application in
several fields, and specifically, recent important relevance in condition monitoring of

machines. Acoustic Emission is defined as a radiation of mechanical elastic waves produced
by the dynamic local rearrangement of the material internal structure. This phenomenon is
associated with cracking, leaking and other physical processes and was described for the
first time by Josef Kaiser in 1950. He described the fact that no relevant acoustic emission
was detected until the pressure applied over the material under test surpassed the
previously highest level applied.

a)
b)
Fig. 5. Examples of common-mode current discharges, a) individual discharge, b) a set of
discharges.
Acoustic Emissions Technique is classified as a passive technique because the object under
test generates the sound and the Acoustic Emission sensor captures it. By contrast, Active
methods rely on signal injection into the system and analysis of variations of the injected
signal due to system interaction. Then an acoustic emission sensor captures the transient
elastic waves produced by cracking or interaction between two surfaces in relative motion
and converts their mechanical displacement into an electrical signal. This waves travel
through the material in longitudinal, transverse (shear) or surface (Rayleigh) waves, but the
majority of sensors are calibrated to receive longitudinal waves. Wherever the crack is

placed, the signal generated travels from the point of fracture to the surface of the material.
The transmission pattern will be affected by the type of material crossed and then isotropic
material will lead to spherical wave front types of propagation only affected by material
surfaces or changes, where the Snell law rules their reflection and reflexion. On Figures 6
and 7 is shown the evolution of acoustic waves inside a Material. On figure 6 it is shown
how reflections on waves due to the defect appear.

Fig. 6. Acoustic Emission Wave Propagation

Fig. 7. Acoustic Emission Wave Propagation in fractured Material
The biggest advantage of this method is probably that it is capable of detecting the earliest
cracks of the system and their posterior growth, making possible fault detection before any
other current method. The main drawback is that it requires additional transducers and a
well controlled environment.
3. Experimental results
Next, the experimental test bench and acquisition system, as well as the results obtained by
each of the presented fault indicators are shown, finally, two inference methods are
presented to merge the obtained information.

3.1 Experimental setup
The test rig used during this research work consists of four ABB M2AA 1.1kW induction
motors, three of them with the drive-end ball bearings under test (with different bearing
fault degradation level), and the other one used to regulate the applied load. Both driving
and loading motors were controlled using independent inverters. Motors under test have
also a cable attached to the drive-end bearings housing with the other side connected to
ground (a hole was mechanized in order to pass the cable through the motor shield),
allowing a low resistance path for CMC acquisition proposes.
The three motors under test have SKF 6205 bearings with normal clearance and nine balls
with diameter of 7.9 mm and pitch of 38.5 mm, and a contact angle of 0.66 radians. The
bearings set under test (labeled healthy, lightly and heavily damaged), is composed by a
healthy one (with very similar vibration levels to other new units tested in previous works)
and other two units with different levels of damage due their operation hours, qualitatively
evaluated with a shock pulse tester from SPM Instruments.

Fig. 8. Experimental test bench and acquisition system scheme.
Regarding the acquisition system, it is based on four different sensors connected to a main
acquisition device. A triaxial shear design MMF branded piezoelectric accelerometer model
KS943B.100 with IEPE (Integrated Electronics Piezo Electric) standard output and linear
frequency response from 0.5 Hz to 22 kHz, was attached using stud mounting to the drive-
end motor end-shield and its data was collected at 20kS/s during 1 second for each
measurement. Phase stator currents were acquired using Hall effect Tektronix A622 probes
with a frequency range from DC to 100 kHz and collected at 20 kHz during 1 second for
each measurement. High frequency CMC signal was measured at the cable attached to the
bearings housing with a Tektronix TCPA300 amplifier and TCP303 current probe, which

provides up to 15 MHz of frequency range, and acquired at 50 MHz during 100 ms for each
measurement. Acoustic emissions were acquired with the use of a Vallen-Systeme GmbH VS-
150M sensor unit with a range from 100 kHz to 450 kHz and resonant at 150 kHz. A Vallen-
Systeme GmbH AEP4 40dB preamplifier was used before data acquisition at a sampling
frequency of 25MS/s during 20ms each measurement. All the described sensors are
connected to a PXI acquisition system from National Instruments formed by different specific
boards.
3.2 Experimental results
3.2.1 Vibrations
The vibration signal RMS contributes clearly to bearings diagnosis. Figures 9, 10 and 11
show the evolution of the RMS value of each motor vibration signals for different speeds
and load patterns tested. Clearly, the healthy motor, in figure 9, shows lower RMS values of
vibration in comparison with the other two units. Figure 11, corresponding to the unit which
was in the worst operational condition according to the SPM measurements performed,
provide also the highest levels of RMS vibration values.

Fig. 9. RMS vibration for healthy unit, all speeds in rpm and loads in percentage of the
nominal one.

Fig. 10. RMS vibration for lightly damaged unit, all speeds in rpm and loads in percentage
of the nominal one.


Fig. 11. RMS vibration for heavily damaged unit, all speeds in rpm and loads in percentage
of the nominal one.
3.2.2 Stator currents
The figure 12a shows an example of stator-phase current in frequency domain over healthy
test bench condition. The stator phase current characteristics bearing fault frequencies are
related with the bearing construction parameters and the equations from (1) to (4) for m = 1
and 2 that are normally used (Obaid etal., 2003). These fault frequencies are not present
along the frequency axis. The fault indicators thresholds for the stator phase current

a)
b)
Fig. 12. Stator current frequency spectrum, from 0 to 500Hz, a) healthy bearings b) fault
bearing

characteristic bearing fault frequencies can be fixed at 5% of the fundamental frequency
amplitude, which is a demanding threshold for diagnosis proposes (Schoen et al., 1995). If
the amplitude of these characteristic fault frequencies surpass the thresholds, imply that it
can be diagnosed clearly the localized bearing fault related, but if this threshold is not
surpassed for any characteristic frequency, it cannot be deduced that bearings are healthy
(Zhou et al., 2009), maybe a generalized bearing defect or a non detectable single defect is
occurring, then, the sidebands of the stator current fundamental harmonic will be analyzed
as general eccentricity fault indicator (Bellini et al., 2008). The stator phase current spectra of
a degraded bearings shows, at figure 10b, sidebands fault frequencies greater than 5% of
fundamental amplitude, but there are not the characteristic bearing fault frequencies. This
effect can be due to eccentricity between rotor and stator for different reasons, so it is
necessary additional features in order to distinguish between eccentricity due to bearings
degradation or due to other fault in the motor.
Regarding the other stator current feature presented, in order to avoid the influence of the
main harmonic power value in the stator current RMS measurement, the acquired signals
have been previously filtered using a band-rejection 5th order Butterworth filter centred in
the power supply main harmonic with a bandwidth of 20 Hz between higher and lower cut-
off frequencies. Tables 1 and 2 compare the RMS filtered values of the heavily and lightly
damaged units with the healthy one.

Heavily Damaged-Healthy ([A] RMS)
Speed [rpm]
Load
[% of nominal torque]
300 750 1050 1500
0 0,004 -0,006 -0,008 -0,007
50 0,036 0,03 0,073 0,044
100 0,018 0,026 0,024 0,024
Table 1. Difference in RMS filtered current value between heavily damaged unit and healthy
one used as reference.

Lightly Damaged-Healthy ([A] RMS)
Speed [rpm]
Load
[% of nominal torque]
300 750 1050 1500
0 0,008 0,002 -0,003 -0,003
50 0,002 -0,011 -0,002 -0,005
100 0,02 0,012 0,003 0,014
Table 2. Difference in RMS filtered current value between lightly damaged unit and healthy
one used as reference.
A significant difference can be clearly appreciated when the motor is heavily damaged
under load condition. Light damage is noticeable under nominal load conditions but its
detection does not seem to be easily reliable.

3.2.3 High frequency bearings pulses
Bearings pulses threshold analysis has been executed to validate theories of correlation
between bearings state (wear, lubrication, distributed defects, etc.) and pulses discharge
over a threshold value. As it can be seen in figure 13 the stator CMC temporal analysis
shows a decrement in the number of pulses surpassing a predefined threshold. The
threshold value is fixed at 75% of the CMC pulse maximal amplitude in healthy cases. A
number of counted pulses less than 75% of counted pulses in healthy bearings, will be the
fault indicator threshold used to distinguish between healthy and degraded bearings.

a)

b)

Fig. 13. Example of common mode current signal acquisition, a) healthy bearings b) fault
bearing.

The results summarized in figure 14, show that over a defined threshold level healthy
bearings undergo a bigger number in comparison to the damaged units. It is noticeable also
that this method is able to detect failure at its initial stage if the threshold is correctly placed.

Fig. 14. Number of bearing pulses over threshold value for all motors under test. Healthy,
lightly damaged and heavily damaged.
3.2.4 Acoustic Emission testing
Acoustic Emission acquired data has been statistically classified by means of value binning
tools and histogram presentation. Fifteen sets of data were acquired for each motor and
averaged. Figure 15 shows the results comparing the RMS voltage values acquired for the
different units under test.

Fig. 15. Acoustic Emission voltage values classification
It is advisable that pulses over 8 V only appeared during the damaged motor testing while
under 7 V that unit does not show more activity than the healthy and lightly damaged units.
Then, the fuzzy inference system designed uses as reference the number of pulses that
surpass the 7 V value, which is the zone where the distinction of the fault severity of the unit
seemed to be more noticeable.

3.3 Inference tools
3.3.1 Look-up tables
A look-up table is a common tool applied in diagnosis field. It contents basically a set of
simple association rules applied over obtained data. The operation consists in analyze a
given combination of inputs in order to select one of the outputs. In the diagnosis field, this
kind of inference tool is as a set of if..then rules collected in a table.
A proposed look-up table is shown in table 3, where a set of features, from the previously
explained have been selected to generate an improved bearings diagnosis system.

FTF harmonic
amplitude
BSF
harmonic
amplitude
BPFO
harmonic
amplitude
BPFI
harmonic
amplitude
Fundamental
sidebands
amplitude
Number
of pulses
over the
threshold
Diagnosis
>5% of
fundamental
Not
necessary
Not
necessary
Not necessary
Not
necessary
< 75%
Bearing cage
fault
(Localized
defect)
Not necessary
>5% of
fundamental
Not
necessary
Not necessary
Not
necessary
< 75%
Bearing ball
fault
(Localized
defect)
Not necessary
Not
necessary
>5% of
fundamental
Not necessary
Not
necessary
< 75%
Bearing
outer race
fault
(Localized
defect)
Not necessary
Not
necessary
Not
necessary
>5% of
fundamental
Not
necessary
< 75%
Bearing
inner race
fault
(Localized
defect)
<5% of
fundamental
<5% of
fundamental
<5% of
fundamental
<5% of
fundamental
>5% of
fundamental
< 75%
Bearing
degradation
(Distributed
or non-
detectable
localized
defect)
<5% of
fundamental
<5% of
fundamental
<5% of
fundamental
<5% of
fundamental
>5% of
fundamental
< 75%
Eccentricity,
but not for
bearing
degradation
Table 3. Look-up table considering single-point stator current characteristic harmonics,
stator current fundamental frequency sidebands evaluation, and number of common mode
pulses.
3.3.2 Fuzzy logic
Fuzzy logic is a useful tool in order to implement reasoning that is ambiguous or imprecise.
In condition monitoring field, the implementation of tolerant and flexible rules is a more
realistic way to generate a diagnosis than the use of crisp and categorical relations.

The analysis of the actual bearing status has been performed using a fuzzy logic inference
implementation (Lou et al., 2004; Ballal et al., 2007), which maps given inputs (in this case
current and vibration RMS values) to a single output, the different signals acquired are
linked to a damage value scaled from 1 to 3.

Fig. 16. Membership function plot for Current RMS. (motor speed: 1500 rpm, motor load: 0%).

Fig. 17. Plotted surface showing the relationship between the system inputs Vibrations RMS
value (g) and Stator Currents RMS value (A) versus the Failure Level output. (Motor speed:
1500 rpm, motor load: 0%)

Unit Matches Success %
Healthy 15 100 %
Lightly Damaged 14 93,33%
Heavy Damage 13 86,66%
Table 4. System testing results.
The membership functions, like figure 16, have been obtained through training and
validation process, for each signal under analysis using real motor data. MATLAB
Adaptive neuro-fuzzy inference system tool has been used for this purpose. Figure 17

shows the obtained relationship between Vibration and Stator Current RMS values against
the Failure Level output for a motor speed of 1500 rpm and a load of 0%.
To perform the evaluation of the monitoring system designed, fifteen sets of data were
collected from the same units and processed. Table 4 summarizes the obtained results.
All healthy data sets were correctly identified, whilst one of the lightly damaged was
recognised as a heavily damaged set and two of the heavily damaged sets were identified as
lightly damaged ones. The percentage of success was reasonably high and its improvement
is still possible if more data sets are used during the system training stage.
4. Conclusions
This chapter tries to offer another point of view in the generation of diagnosis systems and
the use of vibration signal analysis for machine condition monitoring. It has been presented
an overview of multisensory inference approaches used to characterize motor ball bearings,
and their application to a set of motors with distributed fault failure. The results show that a
multivariable design contributes positively to damage monitoring of bearings, being a more
solid solution than just using any of the single signals involved, which can be affected not
only by external disturbances, but also by its own diagnosis limitations, especially dealing
with damage severity evaluation. The selection and fusion of different fault indicators from
different physical magnitudes has been solved by two examples: the application of simple
look-up tables, and the development of a fuzzy system. In both proposed solutions, the
bearings diagnosis reaches an important detection capability, including the possibility to
detect different kinds of bearings faults and/or different levels of fault.
5. References
Ballal S., Khan Z. J., Suryawanshi H. M. & Sonolikar R. L., (2007). Adaptive Neural Fuzzy
Inference System for the Detection of Inter-Turn Insulation and Bearing Wear Faults in
Induction Motor, IEEE Transactions on Industrial Electronics, vol. 54, no. 1, pp. 250-258.
Basak D., Tiwari A. & Das S. P., (2006). Fault diagnosis and condition monitoring od
electrical machines a review, IEEE International Conference on Industrial Technology,
pp. 3061-3066.
Bellini A., Filippetti F., Tassoni C. & Capolino G.-A., (2008). Advances in Diagnostic
Techniques for Induction Machines," IEEE Transactions on Industrial Electronics, vol.
55, no. 12, pp. 4109-4126.
Binder A. & Muetze A., (2008). Scaling Effects of Inverter-Induced Bearing Currents in AC
Machines, IEEE Transactions on Industrial Applications, vol. 40, no. 3, pp. 769-776.
Cabanas M. F., Melero M. G., Orcajo G. A., Cano Rodrguez J. M. & Juan Solares Sariego,
(1996). Tcnicas para el Mantenimiento y diagnstico de Mquinas Elctricas
Rotativas, Marcombo, ISBN: 8426711669.
Cusido J., Delgado M., Navarro L., Sala V.M. & Romeral L., (2010). EMA fault detection
using fuzzy inference tools, IEEE AUTOTESTCON, pp.1-6.
Cusido J., Garcia A., Navarro L. M., Delgado M., Romeral L. & Ortega A., (2009). On-line
measurement device to detect bearing faults on electric motors, IEEE
Instrumentation and Measurement Technology Conference, pp.749-752.
Cusido J., Romeral L., Delgado M., Garcia A., & Ortega J. A., (2007a). Induction machines
fault simulation based on FEM modelling, IEEE European Conference on Power
Electronics and Applications, pp.1-8.

Cusido J., Rosero J., Aldabas E., Ortega J.A. & Romeral L., (2005). Fault detection techniques
for induction motors, IEEE Compatibility in Power Electronics, pp. 85- 90.
Cusido L. R., Garcia A., Rosero J. A. & Ortega J. A., (2007b). Fault detection in induction
machines by using continuous and discrete wavelet decomposition, IEEE European
Conference on Power Electronics and Applications, pp.1-8.
Delgado M., Garcia A., Ortega J. A., Urresty J. & Riba J. R., (2009). Bearing diagnosis
methodologies by means of Common Mode Current, European Conference on Power
Electronics and Applications, pp.1-10, 8-10.
El Hachemi Benbouzid M., (2000). A review of induction motors signature analysis as a
medium for faults detection, IEEE Transactions on Industrial Electronics, vol. 47, no.
5, pp. 984-993.
Frosini L., Bassi E., Fazzi A. & Gazzaniga C., (2008). Use of the stator current for condition
monitoring of bearings in induction motors, Electrical Machines, International
Conference on, pp. 1-6.
Kar C. and Mohanty A. R., (2008). Vibration and current transient monitoring for gearbox
fault detection using multiresolution Fourier transform, Elsevier Journal of Sound and
Vibration, vol. 311, no. 1-2, pp. 109- 132.
Lou X., Kenneth K. A. & Loparo A., (2004). Bearing fault diagnosis based on wavelet
transform and fuzzy inference, Mechanical Systems and Signal Processing, vol. 18, no.
5, pp. 1077-1095.
Muetze A. & Binder A., (2007). Calculation of Motor Capacitances for Prediction of the
Voltage Across the Bearings in Machines of Inverter-Based Drive Systems, IEEE
Transactions on Industry Applications, vol. 43, no. 3, pp. 665-672.
Muetze A. & Binder A., (2007). Calculation of Circulating Bearing Currents in Machines of
Inverter-Based Drive Systems, IEEE Transactions on Industrial Electronics, vol. 54, no.
2, pp. 932-938.
ODonell P., (1986). Report of large motor reliability survey of industrial and commercial
installations, IEEE Transactions on Industrial Applications, vol. IA-21, no. 4.
Obaid R. R., Habetler T. G. & Stack J. R., (2003). Stator current analysis for bearing damage
detection in induction motors, Symposium on Diagnostics for Electric Machines, Power
Electronics and Drives, pp. 182-187.
Obaid, R. R, Habetler, T. G. & Stack, J. R. (2003). Stator current analysis for bearing damage
detection in induction motors, IEEE Proceeding on Power Electronics and Drives,
Symposium on Diagnostics for Electric Machines, pp. 182-187.
Riley C. M., Lin B. K., Habetler T. G. & Kliman G. B., (1999). Stator current harmonics and their
causal vibrations: a preliminary investigation of sensorless vibration monitoring
applications, IEEE Transactions on Industry Applications, vol. 35, no. 1, pp. 94-99.
Schoen R. R., Habetler T. G., Kamran F. & Bartfield R. G., (1995). Motor bearing damage
detection using stator current monitoring, IEEE Transactions on Industry
Applications, vol. 31, no. 6, pp. 1274-1279.
Singh G. K. & Saad Ahmed Saleh Al Kazzaz, (2003). Induction machine drive condition
monitoring and diagnostic research-a survey, Electric Power Systems Research, vol.
64, no. 2, pp. 145-158.
Stack J. R., Habetler T. G. & Harley R. G., (2004). Fault Classification and Fault Signature
Production for Rolling Element Bearings in Electric Machines, Symposium on
Diagnostics for Electric Machines, Power Electronics and Drives, vol. 40, no. 3, pp. 735-739.
Zhou W., Lu B., Habetler T. G. & Harley R. G., (2009). Incipient Bearing Fault Detection via
Motor Stator Current Noise Cancellation Using Wiener Filter, IEEE Transactions on
Industry Applications, vol. 45, no. 4, pp.1309-1317.
14
Vibration Analysis of an Oil Production Platform
Submitted to Dynamic Actions Induced by
Mechanical Equipment
Jos Guilherme Santos da Silva, Ana Cristina Castro Fontenla Sieira,
Luciano Rodrigues Ornelas de Lima and Bruno Dias Rimola
State University of Rio de Janeiro (UERJ)
Brazil
1. Introduction
Structural engineers have long been trying to develop solutions using the full potential of its
composite materials. At this point there is no doubt that the structural solution progress is
directly related to an increase in materials science knowledge.
On the other hand, the competitive trends of the world market have long been forcing
structural engineers to develop minimum weight and labour cost solutions. A direct
consequence of this new trend design is a considerable increase in problems related to
unwanted floor vibrations. For this reason, the structural floors systems can become
vulnerable to excessive vibrations, for example, produced by impacts such as mechanical
equipment (rotating machinery) (Rimola, 2010; Rimola, 2010a; Rimola, 2010b).
This way, the present paper investigated the dynamic behaviour of an oil production
platform made of steel and located in Santos basin, So Paulo, Brazil. The structural model
consists of two steel decks with a total area of 1915 m
2
(upper deck: 445 m
2
and lower deck:
1470 m
2
), supported by vertical sections made of tubular steel members (steel jacket), and
piled into the seabed. A variety of mechanical equipment was located on the steel decks of
the structural model, related to electrical generators and compressors (Rimola, 2010).
The soil representation was based on the Winklers Theory (Winkler, 1867). This theory
simulates the soil behaviour as a group of independent springs, governed by the linear-
elastic model. In the Winklers model, the soil stiffness was considered as the necessary
pressure to produce a unitary displacement (Winkler, 1867).
The proposed computational model, developed for the oil production platform dynamic
analysis, adopted the usual mesh refinement techniques present in finite element method
simulations implemented in the GTSTRUDL program (GTSTRUDL, 2009). In this finite
element model, floor steel girders and columns were represented by three-dimensional
beam elements, where flexural and torsion effects were considered. The steel decks were
represented by shell finite elements. In this investigation, it was considered that both
structural elements (steel beams and steel deck plates) have total interaction with an elastic
behaviour.
The structural model dynamic response was determined through an analysis of its natural
frequencies and peak accelerations. The results of the dynamic analysis were obtained from


282
an extensive numerical study, based on the finite element method using the GTSTRUDL
program (GTSTRUDL, 2009). In this investigation, dynamic loadings coming from the
rotating machinery (electrical generators and compressors) were applied on the steel decks
of the structural system (production platform).
A numerical analysis was performed, in order to obtain the dynamic impacts on the deck
structure coming from the electrical generators and compressors. Based on the peak
acceleration values, obtained on the structure steady-state response, it was possible to
evaluate the structural model performance in terms of human comfort, maximum tolerances
of the mechanical equipment and vibration serviceability limit states of the structural
system, based on the design code recommendations (CEB 209/91, 1991; ISO 1940-1, 2003;
ISO 2631-1, 1997; ISO 2631-2, 1989; Murray et al., 2003).
2. Vibration analysis of steel floors
Besides the evaluation of the structural systems behaviour when submitted to dynamic
loads, the causes and effects of vibration on people have been subject of many studies and
experiments, due to the fact that they affect people in different ways, causing discomfort,
health problems, reduced ability concentration and efficiency at work or sickness, in the case
of vibrations of very low frequencies.
(Reiher & Meister, 1946) developed a scale used to determine levels of acceptable vibration
in floors. This scale was developed based on experiments in which a group of people was
submitted to vibration, whose frequency varied from 1 Hz to 100 Hz. According to this
scale, the vibration levels can be classified into several levels, depending on the amplitude
and frequency.
(Srinivasulu & Vaidyanathan, 1976) presented the principles of analysis, design and
construction of machines of different types. The authors investigated several factors to be
considered during the design of machine foundations, in order to obtain the best solution,
leading to better operation and reduce the undesirable effects of vibrations on the structure.
(Bachmann & Ammann, 1987) have studied the necessary procedures for the analysis of
structures under dynamic loads coming from machines, including machines with rotating
parts. The authors treat from the load formulation, also dealing with the effects of
machinery induced vibrations in structures and measures to avoid such a problem. In their
work are also included the acceptance criteria, both from the point of view of the structure
and the point of view of human comfort.
(Griffin, 1996) indicates some reasons to measure human exposure to vibration, especially
the following: development of standardized documentation on vibrations in the human
body; determination of vibration levels and reduction in frequency range that can prejudice
the human body and providing data that may be used for comparison between two or more
occupational environments.
(Lenzen, 1996) observed during the development of his research, that the scale developed by
(Reiher & Meister, 1946) did not take into account the influence of damping on the human
perception of vibration. Through his studies in the laboratory, it was modified the scale of
(Reiher & Meister, 1946). This modified scale showed satisfactory results in floors with
damping ratios of up to 5%.
According to (Vasconcelos, 1998), establishing the concept of human vibration discomfort
can be a difficult task. There are several factors that can influence the subjective feeling of
discomfort, such as the socio-cultural people, the type of activity performed, the persons
Vibration Analysis of an Oil Production
Platform Submitted to Dynamic Actions Induced by Mechanical Equipment

283
psychological state at the time of the event, environmental factors, noise, etc. It is not easy to
simulate these conditions in the laboratory to reduce the variability of individual responses.
Thus, the limit of comfort of people subjected to vibration can be regarded as a rather
subjective measure, generating some controversy as to the acceptable values of accelerations
imposed.
(Zhou & Shi, 2001) considered that the elimination of vibration of rotating machinery is an
important engineering problem. In their study, they presented a detailed review of the
developed research that deals with the active balancing of rotors in real time and active
control of vibration of rotating machinery, as well as dynamic modelling and analysis
techniques for rotating systems. The authors report that the major problem found by the
scheme of active control of vibration is the limited number of actuators to control an
unlimited number of vibration modes.
(Pereira, 2005) presents a study on the vibration related to human comfort and perception,
focusing on the suitability of buildings for vibration levels, aiming the generation of curves
related to the perception and human comfort and vibration by means of laboratory
experiments and comparing the results to the limits of vibration to other investigations and
the design codes (ISO 2631-2, 1989).
The experimental tests developed by (Pereira, 2005) considered 30 volunteers, 15 men and 15
women exposed to vertical vibration at a frequency band ranging from 12 to 80 Hz in sitting
and standing posture. The author also performed an analysis on the uncertainty of the
outcome of the limit of perception, verifying the existence of a range of vibration in which
individuals are not sure whether or not they are able to detect the vibration. It also aimed to
know the vertical vibration levels that people find uncomfortable at their home environment,
to determine the relationship between the perception threshold and comfort. According to
these results, it was proved that the reduction in amplitude of the movement to higher
frequency vibration becomes more difficult to be detected, reducing the sensitivity of people.
(Milet, 2006) discusses the basic concepts of dynamic analysis of machine foundations,
investigating some analytical strategies and numerical methods available for designing. In
this work, some design recommendations were presented and discussed.
(Souza et al., 2007) developed a prototype that allows, looking through a simple system,
based on an unbalanced rotor, possible structural data caused by the resonance
phenomenon, also allowing comparisons to be made with more complex structural systems.
Furthermore, the experiment presented as being practical and simple, can serve as an
analytical tool in the classroom, thus giving a better understanding of phenomena related to
the vibration system.
(Assuno, 2009) addressed the issues and the most important conditions for a dynamic
analysis of elevated frame structures, where equipment were allocated for industrial
processes. The author developed a study related to the main causes of dynamic actions
coming from industrial equipment and examined a framed structure supporting an
unbalanced machine. Through this investigation, the author demonstrated that the
developed computational model was appropriate to simulate the transmission of efforts and
contribution of the vibrating mass on the responses of the framed structures.
3. Loads generated by mechanical equipment (rotating machinery)
Knowledge about the dynamic behaviour of rotors of rotating machinery, even in the design
phase, has become an increasingly crucial factor, considering that it is not desirable to take


284
corrective actions after the beginning of activities. Material costs and delivery of services are
relatively high when compared with the profits increasingly reduced according to the rules
imposed by the market and the fact that such corrective actions still involve a period in
which the equipment will need to become inoperative, which means losses, because it will
not generate any profit for the period.
According to (Dias Junior, 2009), among several factors that contribute to the transmission of
rotational energy to the vibratory movements of the machine, the well-known is
undoubtedly the unbalance of the rotor. The rotor is the rotating part of a machine or engine
which may be coupled elements as disks, generators, gears, etc. According to (Dias Junior,
2009), due to the unbalance, the force that acts at the rotor gravity centre, pull the shaft out
of the line joining the two bearings, forcing the shaft to rotate stressed. This movement is
called precession movement.
Rotors are supported on bearings, which are the elements responsible for connecting the
movable and fixed structure of a rotating machine. In addition to this point, to absorbing
energy, another function of the bearings is to guide or restrict movement during the rotation
axis (Silva, 2004).
The process of balancing a rotor is a key factor to minimize the vibrations generated by
electric motors. Depending on the vibration level of these engines, the structural system that
supports the equipment can be compromised by fatigue or even premature failure. The
balancing process is intended to improve the distribution of mass of a body, so that by
turning around their bearings, produces no unbalance forces, keeping the vibrations and
dynamic loads within suitable limits for the machine operation.
The balance can be achieved up to a certain limit, since after this process the rotor still
presents imperfection in the mass distribution, called residual unbalance. It is worth
mentioning that there is a direct relationship between the residual unbalance and vibration
level of the machine, which depends on many factors (mass housing and the foundation,
stiffness of the bearings and foundation, occurrence of resonances etc.). Anyway, there are
allowable levels of residual unbalance, consistent with good practice of machine design.
3.1 Excitation forces: Unbalanced mass
Unbalanced mass is defined as a mass located at a distance d measured from the geometric
centre of the shaft. The mass remains in a plane perpendicular to the axis y and it is a
constant coordinate, as illustrated in Figure 1.
Based on Figure 1, it can be deduced that the force caused by unbalanced mass, acting on
the axis, according to directions in the shown coordinate system, can be written as presented
in Equations (1) and (2). In Equations (1) and (2), the dynamic forces generated by
unbalanced mass have a frequency similar to the rotational frequency of the axis
( )
2
u u
F m d sin t = (1)
( )
2
w u
F m d cos t = (2)
3.2 Unbalance quantification
As previously mentioned, the unbalance is characterized by a mass located at a certain
distance from the rotation axis, see Figure 1. Therefore, the unbalance is always measured
by a product mass x distance, see Equation (3). The rotor must be subjected to a balancing

285
procedure, in order to achieve a tolerable minimum. This value is called the Permissible
Residual Unbalance and designated by the symbol U.
U mass distance = (3)

Fig. 1. Unbalanced mass (Lpez, 2002)
It was observed in Equation (3), that the allowable residual unbalance is directly
proportional to the mass of the rotor, i.e., as heavier was the rotor, greater it will be the
permissible residual unbalance. It is appropriate to relate the allowable residual unbalance,
U, and the rotor mass, m, in terms of the Specific Allowable Residual Unbalance, as shown
in Equation (4).

U
e
m
= (4)
As larger is the rotation speed, smaller should be the residual unbalance, since the
centrifugal force, F
cent
, increases with the square of the speed of it, as shown in Equation (5).
In Equation (5) the centrifugal force, F
cent
is expressed in N.

2
cent
F m e = (5)
Based on many years of experience, experts decided that the product of the angular velocity
of rotor rotation and the specific allowable residual unbalance must be constant, i.e., to
increase the speed of rotation it is necessary to reduce the specific residual unbalance.
This product is called Balance Quality Grade and it is designated by the letter G, see Table 1.
To find a wide variety of existing rotors it was necessary to assign, depending on the type of
rotor and its application, a value for this constant, see Table 1. Table 1 reproduces the G
values which deals with quality of balancing rotating rigid bodies (ISO 1940-1, 2003).
3.3 Determination of unbalanced forces
As mentioned before, the unbalance of the rotor produces a dynamic load that depends on
the mass, the equipment angular velocity and the eccentricity between the equipment


286
G
e x
(mm/s)
Rotor Types - General Examples
G 4000 4000
Crankshaft/drives of rigidly mounted slow marine diesel engines with
uneven number of cylinders.
G 1600 1600 Crankshaft/drives of rigidly mounted large two-cycle engines.
G 630 630
Crankshaft/drives of rigidly mounted large four-cycle engines;
Crankshaft/drives of elastically mounted marine diesel engines.
G 250 250 Crankshaft/drives of rigidly mounted fast four-cylinder diesel engines.
G 100 100
Crankshaft/drives of fast diesel engines with six or more cylinders;
Complete engines (gasoline or diesel) for cars, trucks and locomotives.
G 40 40
Car wheels, wheel rims, wheel sets, drive shafts; Crankshaft/drives of
elastically mounted fast four-cycle engines with six or more cylinders;
Crankshaft/drives of engines of cars, trucks and locomotives.
G 16 16
Drive shafts (propeller shafts, cardan shafts) with special requirements and
parts of crushing machines; Individual components of engines (gasoline or
diesel) for cars, trucks and locomotives; Crankshaft/drives of engines with
six or more cylinders under special requirements.
G 6.3 6.3
Parts of process plant machines; Marine main turbine gears (merchant
service); Centrifuge drums; Paper machinery rolls; print rolls; Fans;
Assembled aircraft gas turbine rotors; Flywheels; Pump impellers;
Machine-tool and general machinery parts; Medium and large electric
armatures (electric motors having at least 80 mm shaft height) without
special requirements; Small electric armatures, often mass produced, in
vibration insensitive applications and/or with vibration-isolating
mountings; Individual components of engines under special requirements.
G 2.5 2.5
Gas and steam turbines, including marine main turbines (merchant
service); Rigid turbo-generator rotors; Computer memory drums and discs;
Turbo-compressors; Machine-tool drives; Medium and large electric
armatures with special requirements; Small electric armatures not
qualifying for one or both of the conditions specified for small electric
armatures of balance quality grade G 6.3; Turbine-driven pumps.
G 1 1
Tape recorder and phonograph (gramophone) drives; Grinding-machine
drives; Small electric armatures with special requirements.
G 0.4 0.1 Spindles, discs and armatures of precision grinders; Gyroscopes.
Table 1. Balance quality values (ISO 1940-1, 2003)
gravity center and the rotation axis. In sequence, Equation (6) determines the dynamic load
amplitude generated by the unbalance of a rotor, as follows:
( )
2
0
P m R m R = = (6)

287
Where:
P
0
: dynamic loading amplitude;
M : total mass in rotation;
: equipment frequency;
R. = G : equipment balance quality grade, see Table 1: (ISO 1940-1, 2003).
For rotors of electric motors, R. was considered equal to 0.0025 m/s. Substituting this
value in Equation (6), it is obtained:
( )
0
P m 0.0025 = (7)
Considering an unbalanced load spinning around an axis, the procedure for obtaining the
global dynamic force acting on a plane is to apply the force in two orthogonal directions.
One of these forces is applied in the horizontal direction with the angle phase equal to zero
and the other in vertical direction with the angle phase equal to 1/4 of the period of
vibration of unbalanced force.
Thus, as time increases, there is a variation of the two forces so that the composition
(horizontal and vertical directions) results in a harmonic unbalanced force where one
component will be multiplied by sin (t) and the other by sin (t + /2). This way, when
one harmonic component presents maximum values the other one is equal to zero and vice
versa. The value of the dynamic force is obtained by the vector sum of the components in
the vertical and horizontal directions as presented in Equation (8).

0 0
P(t) P sin( t) P sin( t )
2
= + + (8)
3.3 Dynamic loading modelling
To perform the numerical analysis of the oil production platform developed in this
investigation, it was used the data in accordance with Table 2. In sequence, Figure 2 shows
the design of the equipment.

Equipment data
Protective cover 1.2 kN
Coupling 5.3 kN
Gear unit 37.5 kN
Motor swing 15 kN
Rotor weight 10.8 kN
Input frequency 30 Hz
Output frequency 0.94 Hz
Table 2. Equipment data
The dynamic load modelling considered two components related to vertical and horizontal
directions. Table 3 shows the dynamic loads applied on the structural system steel deck.
These actions were properly combined in order to better represent the dynamic excitation
induced by equipment on the structure.


288

Fig. 2. Driving unit (motor, coupling and gear) supported by a steel beam (Rimola, 2010)

Equipment Weight (kN) Frequency (rad/s) R (m/s) P
0
(kN)
Rotor 10.80 188.49 0.0025 0.51
Gear 18.75 6.03 0.0025 0.028
Table 3. Dynamic actions related to the equipment
4. Investigated structural model
The structural system investigated in this paper is related to an oil production platform
made of steel and located in Santos basin, So Paulo, Brazil. The structural model is
supported by vertical sections made of tubular steel members (steel jacket), piled into the
seabed by steel piles and consists of two steel decks with a total area of 1915 m
2
(upper deck:
445 m
2
and lower deck: 1470 m
2
), see Figures 3 and 4.

Fig. 3. Investigated structural model

289
The structural model is formed by steel beams and columns and steel deck plates, as
presented in Figures 3 and 4. This platform is constituted by a lot of structural elements with
very different geometrical characteristics; see (Rimola, 2010). A 2.05x105MPa Youngs
modulus was adopted for the steel beams, columns and decks.

Fig. 4. Finite element model: oil production platform with steel jacket and piles
Nodes: 1824
Elements: 3588
Beam: 3079
Shell: 509
Degrees of Freedom: 10872


290

Fig. 5. Geotechnical profile
5. Computational modelling
The proposed computational model, developed for the structural system dynamic analysis,
adopted the usual mesh refinement techniques present in finite element method simulations
implemented in the GTSTRUDL program (GTSTRUDL, 2009).

291
In this computational model, floor steel girders and columns were represented by a three-
dimensional beam element with tension, compression, torsion and bending capabilities. The
element has six degrees of freedom at each node: translations in the nodal x, y, and z
directions and rotations about x, y, and z axes. On the other hand, the steel deck plates were
represented by shell finite elements (GTSTRUDL, 2009).
In this investigation, it was considered that both structural elements (steel beams and steel
deck plates) have total interaction with an elastic behaviour. The finite element model has
1824 nodes, 3079 three-dimensional beam elements, 509 shell elements and 10872 degrees of
freedom, as presented in Figure 4.
5.1 Geotechnical data
The data related to soil were obtained by means of three boreholes (Standard Penetration
Tests: SPT) in a depth range varying from 43.5m to 178.3m. The boreholes allowed the
definition of the geotechnical profile (Figueiredo Ferraz, 2004) adopted in the finite element
modelling, see Figure 5. Characterisation and resistance tests performed in laboratory
provided specific weight, friction angle and cohesion values illustrated in Figure 5. In Figure
5,
sub
is the submerged specific weight (kN/m
3
); c represents the soil cohesion in (kN/m
2
)
and is soil friction angle (degree).
5.2 Soil-structure interaction
When the study of half-buried columns is considered, the usual methodology to the
formulation of the soil-structure interaction problem utilizes the reaction coefficient concept,
originally proposed by (Winkler, 1867). In the case of laterally loaded piles, the analysis
procedure based on (Winkler, 1867) is analogue to the one used for shallow foundations, see
Figure 6.
The soil behaviour is simulated by a group of independent springs, governed by a linear-
elastic model. The foundation applies a reaction in the column normal direction and it is
proportional to the column deflection. The spring stiffness, designated by reaction
coefficient (k
h
) is defined as the necessary pressure to produce a unitary displacement
(Winkler, 1867), as presented in Equation (9).

(a) Shallow foundation (b) Laterally loaded pile
Fig. 6. Foundation analysis models


292

h
p
k
y
= (9)
Where:
p : applied pressure, in (N/m
2
);
y : soil deflection, in (m).
(Terzaghi, 1955) considered that the reaction coefficient (k
h
) for piles in cohesive soils (clays),
does not depend on the depth of the pile and may be determined by the following equation.
( )
h s1
0.3048m
k k
1.5d
= (10)
Where:
k
s1
: modulus to a squared plate with a length of 0.3048m (1 ft), in (MN/m
3
);
d : column width (pile), in (m).
Table 4 presents typical values for k
s1
for consolidated clays.

Clays consistency k
s1
(t/ft
3
) k
s1
(MN/m
3
)
Stiff 75 26.0
Very Stiff 150 52.0
Hard 300 104.0
Table 4. Typical values for k
s1
(Terzaghi, 1955)
For piles in non-cohesive soils (sand), it is considered that the horizontal reaction coefficient
(k
h
) varies along the depth, according to Equation (11).

h h
z
k n
d
= (11)
Where:
n
h
: stiffness parameter for non-cohesive soils, in (MN/m
3
);
d : column width (pile), in (m).
Typical values for n
h
obtained by (Terzaghi, 1955), as a function of the sand relative density
under submerged and dry condition, are presented in Table 5.

Relative density
n
h
(dry)
(t/ft
3
)
n
h
(dry)
(MN/m
3
)
n
h
(submerged)
(t/ft
3
)
n
h
(submerged)
(MN/m
3
)
Loose 7 2.4 4 1.4
Medium 21 7.3 14 4.9
Dense 56 19.4 34 11.8
Table 5. Typical values for n
h
(Terzaghi, 1955)
Based on the horizontal reaction coefficients values (k
h
) and the column width (d), the
foundation stiffness parameter (k
0
) is determined by using Equation (12) (Poulos & Davis,
1980):

0 h
k k d = (12)

293
Based on the subsoil geotechnical profile (see Figure 5) and using the analysis procedure
based on the Winkler model (Winkler, 1867) the horizontal reaction coefficients on the piles
(k
h
) were determined as a function of the type of the soil. Applying the horizontal reaction
coefficients (k
h
) on Equation (12), the foundation stiffness parameters values (k
0
) were
calculated.
The foundation stiffness parameters values (k
0
) were used to determine the springs stiffness
(k) placed in the computational model to simulate the soil behaviour. The spring elements
which simulate the piles were discretized based on a range with length equal to 1m (one
meter). For each range of 1m it was placed a translational spring in the transversal direction
of the pile section axis with a stiffness value equal to the value obtained for the horizontal
reaction coefficient evaluated by the Winkler model (Winkler, 1867). Table 6 presents the
springs stiffness coefficients simulating.

Depth (m) Layer Description
Pile Profile Springs stiffness
k (kN/m) (mm) Thickness (mm)
1
Medium sand 2108 55
4850.1163
2 9700.2327
3 14550.3490
4 19400.4653
5 24250.5816
6 29100.6980
7 a 18 Stiff clay 2134 55 5279.6981
19
Fine sand 2134 55
92152.2102
20 97002.3266
21 101852.4429
22 106702.5592
23 a 25 Medium clay 2134 55 281.5839
26
Medium sand 2134 55
126103.0245
27 130953.1408
28 135803.2572
29 140653.3735
30 145503.4898
31 150353.6062
32 155203.7225
33 160053.8388
34 164903.9551
35
Fine sand 2134 55
169754.0715
36 174604.1878
37 179454.3041
38 184304.4204
39 189154.5368
40 194004.6531
41 198854.7694
42 203704.8858
43 208555.0021
44 a 96 Stiff clay 2134 55 5279.6981
Table 6. Stiffness coefficients of soil representative springs


294
5.3 Structural damping modelling
It is called damping the process which energy due to structural system vibration is
dissipated. However, the assessment of structural damping is a complex task that cannot be
determined by the structure geometry, structural elements dimensions and material
damping (Clough & Penzien, 1995).
According to (Chopra, 2007), it is impossible to determine the damping matrix of a
structural system through the damping properties of each element forming the structure
with the same way it is determined the structure stiffness matrix, for example. This is
because unlike the elastic modulus, which is used in the stiffness evaluation, the damping
materials properties are not well established.
Although these properties were known (Chopra, 2007), the resulting damping matrix would
not take into account a significant portion of energy dissipated by friction in the structural
steel connections, opening and closing micro cracks in the concrete, friction between
structure and other elements that are bound to it, such as masonry, partitions, mechanical
equipment, fire protection, etc. Some of these energy dissipation sources are extremely
difficult to identify.
The physical evaluation of the damping of a structure is not considered properly as if their
values are obtained by experimental tests. However, these tests often require time and cost
that in most cases is very high. For this reason, damping is usually achieved in terms of
contribution rates, or rates of modal damping (Clough & Penzien, 1995).
With this in mind, it is common to use the Rayleigh damping matrix, which considers two
main parts, one based on the mass matrix contribution rate () and another on the stiffness
matrix contribution rate (), as can be obtained using Equation (13). It is defined M as the
mass matrix and K as the stiffness matrix of the system (Craig Jr., 1981; Clough & Penzien,
1995; Chopra, 2007).
= + C M K (13)
Equation (13) may be rewritten, in terms of the damping ratio and the circular natural
frequency (rad/s), as presented in Equation (14):

0i
i
0i
2 2

= + (14)
Where:
i
: damping ratio related to the i
th
vibration mode;
0i
: circular natural frequency related to the i
th
vibration mode.
Isolating the coefficients and from Equation (14) and considering the two most
important structural system natural frequencies, Equations (15) and (16) can be written.

i 01 01 01
2 = (15)

( )
2 02 1 01
02 02 01 01
2
(16)
Based on two most important structural system natural frequencies it is possible to calculate
the values of and . In general, the natural frequency
01
is taken as the lowest natural

295
frequency, or the structure fundamental frequency and
02
frequency as the second most
important natural frequency.
In the technical literature, there are several values and data about structural damping. In
fact, these values appear with great variability, which makes their use in structural design
very difficult, especially when some degree of systematization is required. Based on the
wide variety of ways to considering the structural damping in finite element analysis,
which, if used incorrectly, provide results that do not correspond to a real situation, the
design code CEB 209/91 (CEB 209/91, 1991) presents typical rates for viscous damping
related to machinery support of industrial buildings, as shown in Table 7.

Construction Type Minimum Mean Maximum
Reinforced Concrete 0.010 0.017 0.025
Prestressed Concrete 0.007 0.013 0.020
Composite Structures 0.004 0.007 0.012
Steel 0.003 0.005 0.008
Table 7. Typical values of damping ratio for industrial buildings (CEB 209/91, 1991)
Based on these data, it was used a damping coefficient of 0.5% ( = 0.5% or 0.005) in all
modes. This rate takes into account the existence of few elements in the oil production
platform that contribute to the structural damping. Table 8 presents the parameters and
used in the forced vibration analysis to model the structural damping in this investigation.

f
01

(Hz)
f
02

(Hz)
01

(rad/s)
02

(rad/s)

0.674 0.716 4.2364 4.4968 0.02172 0.00115
Table 8. Values of the coefficients and values used in forced vibration analysis
6. Natural frequencies and vibration modes
The production platform natural frequencies were determined with the aid of the numerical
simulations, see Table 9, and the corresponding vibration modes are shown in Figure 7 and
8. Each natural frequency has an associated mode shape and it was observed that the first
vibration modes presented predominance of the steel jacket system.

Natural Frequencies
f
0i

(Hz)
Vibration Modes
f
01
0.67 Mode 1
Vibration modes with predominance of the steel jacket
system.
f
02
0.71 Mode 2
f
03
1.20 Mode 3
f
08
1.99 Mode 8
Vibration modes with predominance of the steel deck
displacements.
f
17
2.61 Mode 17
f
49
4.14 Mode 49
Table 9. Production platform natural frequencies


296
It can be observed in Figure 7, that the three first vibration modes presented predominance
of displacements in the steel jacket system. In the 1
st
vibration mode there is a predominance
of translational displacements towards the y axis in the finite element model. In the 2
nd

vibration mode a predominance of translational effects towards the axis x of the
numerical model was observed. The third vibration mode presented predominance of
torsional effects on the steel jacket system with respect to vertical axis z. Flexural effects
were predominant on the steel deck system and can be seen only from higher order
vibration modes, see Figure 8.
However, flexural effects were predominant in the steel deck plate (upper and lower),
starting from the eighth vibration mode (f
8
= 1.99 Hz - Vibration Mode 8), see Table 9. It is
important to emphasize that torsional effects were present starting from higher mode
shapes, see Table 9. Figures 7 and 8 illustrated the mode shapes corresponding to six natural
frequencies of the investigated structural system.

a) 1
st
vibration mode b) 2
nd
vibration mode c) 21
st
vibration mode
Fig. 7. Vibration modes with predominance of the steel jacket system

297
7. Structural system dynamic response
The present investigation proceeds with the evaluation of the steel platforms performance
in terms of vibration serviceability effects, considering the impacts produced by mechanical
equipment (rotating machinery). This strategy was considered due to the fact that
unbalanced rotors generate vibrations which may damage their components and supports
and produce dynamic actions that could induce the steel deck plate system to reach
unacceptable vibration levels, leading to a violation of the current human comfort criteria
for these specific structures. For this purpose, forced vibration analysis is performed
through using the computational program (GTSTRUDL, 2009). The results of forced
vibration models are obtained in terms of the structural system displacements, velocities
and peak accelerations.
For the structural analysis, it was considered the simultaneous operation of three machines
on the steel deck. The nodes of application of dynamic loads in this situation are shown in
Figure 9. With respect to human comfort, some nodes of the finite element model were
selected near to the equipment in order to evaluate the steel deck dynamic response
(displacements, velocities and accelerations). These nodes are also shown in Figure 9.
The dynamic loading related to the rotor was applied on the nodes 9194, 9197 and 9224 and
the corresponding dynamic loading associated to the gear was applied on the nodes 9193,
9196 and 9223, as presented in Figure 9. The description of the dynamic loadings (rotor and
gear) was previously described on item 3. It should be noted that the positioning of the
machines was based on the equipment arrangement drawings of the investigated platform
(Figueiredo Ferraz, 2004).
The analysis results were compared with limit values from the point of view of the
structure, operation of machinery and human comfort provided by international design
recommendations (CEB 209/91, 1991; ISO 1940-1, 2003; ISO 2631-1, 1997; ISO 2631-2, 1989;
Murray et al., 2003). It must be emphasized that only the structural system steady-state
response was considered in this investigation.
The frequency integration interval used in numerical analysis was equal to 0.01 Hz ( =
0.01 Hz). It was verified that the frequency integration interval simulated conveniently the
dynamic characteristics of the structural system and also the representation of the proposed
dynamic loading (Rimola, 2010).
In sequence of the study, Tables 10 to 12 present the vertical translational displacements,
velocities and accelerations, related to specific locations on the steel deck, near to the
mechanical equipment, see Figure 9, calculated when the combined dynamic loadings (rotor
and gear) were considered.
These values, obtained numerically with the aid of the proposed computational model, were
then compared with the limiting values proposed by design code recommendations (CEB
209/91, 1991; ISO 1940-1, 2003; ISO 2631-1, 1997; ISO 2631-2, 1989; Murray et al., 2003). Once
again, it must be emphasized that only the structural system steady state response was
considered in this investigation.
The allowable amplitudes are generally specified by manufactures of machinery. When
manufactures data doesnt indicate allowable amplitudes, design guides recommendations
(ISO 1940-1, 2003) are used to determine these limiting values for machinery performance,
see Table 10. The maximum amplitude value at the base of the driving support (Node 9197:
see Figure 9), on the platform steel deck was equal to 446 m (or 0.446 mm or 0.0446 cm), see
Table 10, indicating that the recommended limit value was violated and the machinery
performance can be inadequate (0.446 mm > 0.06 mm) (ISO 1940-1, 2003).


298

a) 8
th
vibration mode f
08
=1.99 Hz

b) 17
th
vibration mode f
17
=2.61 Hz

c) 49
th
vibration mode f
49
=4.14 Hz

Fig. 8. Vibration modes with predominance of the steel deck system

299

Fig. 9. Selected nodes for application of the dynamic loads

Rotor Support
(Node: 9194) (m)
Rotor Support
(Node: 9197) (m)
Rotor Support
(Node: 9224) (m)
Amplitude Limit
Values (m)
54 446 7
40 to 60*
Gear Support (Node:
9193) (m)
Gear Support (Node:
9196) (m)
Gear Support (Node:
9223) (m)
77 432 38
* For vertical vibration for high speed machines (>1500 rpm) (ISO 1940-1, 2003).
Table 10. Vertical displacements related to the combined dynamic loading (driving)

Rotor Support
(Node: 9194) (mm/s)
Rotor Support
(Node: 9197) (mm/s)
Rotor Support
(Node: 9224) (mm/s)
Velocity Limit
Values (mm/s)
10.18 84.12 11.89
0.70 to 2.80*
Gear Support (Node:
9193) (mm/s)
Gear Support (Node:
9196) (mm/s)
Gear Support (Node:
9223) (mm/s)
14.49 81.46 7.27
* Tolerable velocity for electrical motors according to (ISO 1940-1, 2003).
Table 11. Velocities related to the combined dynamic loading (driving)
The maximum velocity value calculated at the base of the driving support (Node 9197: see
Figure 9), on the platform steel deck was equal to 84.12 mm/s, see Table 11. The allowable
velocity considering a perfect condition to machinery performance is located in the range of
0.7mm/s to 2.8 mm/s, see (ISO 1940-1, 2003), as presented in Table 11. This velocity value is


300
not in agreement with those proposed by the design codes (84.12 mm/s > 2.8 mm/s) (ISO
1940-1, 2003), violating the recommended limits.

Steel Deck
(Node: 9098)
(m/s
2
)
Steel Deck
(Node: 9137)
(m/s
2
)
Steel Deck
(Node: 9173)
(m/s
2
)
Steel Deck
(Node: 9189)
(m/s
2
)
Acceleration Limit
Values (m/s
2
)
5.62 6.16 4.67 27.33
0.315 to 1.0*
Steel Deck
(Node: 9221)
(m/s
2
)
Steel Deck
(Node: 9282)
(m/s
2
)
Steel Deck
(Node: 9290)
(m/s
2
)
Steel Deck
(Node: 9317)
(m/s
2
)
11.62 8.19 11.77 0.61
* Acceptable acceleration values for human comfort in accordance with (ISO 2631-1, 1997).
Table 12. Accelerations related to the combined dynamic loading (driving)
People working temporarily near to driving could be affected in various degrees (human
discomfort). The allowable acceleration value when the human comfort is considered (ISO
2631-1, 1997) is located in the range of 0.315 m/s to 1.0 m/s, as illustrated in Table 12. The
peak acceleration value calculated on the platform steel deck was equal to 27.33 m/s, see
Table 12. This maximum acceleration value violated the recommended limits proposed by
the design codes (27.33 m/s
2
> 1.0 m/s
2
) (ISO 2631-1, 1997), causing human discomfort.
Based on the numerical results obtained with the present investigation, it was clearly
demonstrated that the investigated structural model presented problems due to excessive
vibration. This way, some changes on the horizontal steel bracing system were proposed by
the authors, in order to reduce the vibration effects.
This way, Figure 10 shows the proposed steel bracing members which were added to the
original design of the investigated structural model. The central idea was to propose a
bracing system that produces an increasing in the steel deck stiffness and, consequently, can
cause a reduction of the steel deck system dynamic response, when submitted to the rotor
and gear dynamic loads.
In sequence, Figure 11 shows the vibration modes of the production platform steel deck
system before and after the addition of the proposed bracing members shown in Figure 10.
Comparing the presented vibration modes, it may be concluded that modal amplitudes
have been modified, leading to a reduction of the structural system dynamic response, in
terms of displacements, velocities and accelerations.
Figure 12 shows the response spectra of vertical translational displacements, for the
supports and rotors, and, for particular nodes of the steel deck of the platform. Only three
graphs were presented due to the fact that this figure (Figure 12) represents, in general, the
dynamic response of the system. Based on the structure dynamic response, the peak of
interest for the forced vibration numerical analysis, associated with the excitation frequency
of the equipment (f = 30 Hz) is indicated in the Figure 12.
Tables 13 to 15 present the vertical translational displacements, velocities and accelerations,
related to specific locations on the steel deck, near of the mechanical equipment, see Figure
9, calculated with the introduction of the proposed steel bracing system, when the combined
dynamic loadings were considered.
The values presented in Tables 13 to 15 were obtained numerically with the aid of the
developed finite element model. These values were then compared with the limiting values
proposed by design code recommendations (CEB 209/91, 1991; ISO 1940-1, 2003; ISO 2631-1,

301

a) Lower deck

b) Upper deck

Fig. 10. Steel deck with the introduction of the proposed bracing system
Proposed bracing system
Proposed bracing system


302

a) Before the introduction of the bracing system. Mode: 1203 - Frequency: 30 Hz
b) After the introduction of the bracing system. Mode: 1195 - Frequency: 30 Hz

Fig. 11. Steel deck vibration modes
1997; ISO 2631-2, 1989; Murray et al., 2003), after the introduction of the proposed bracing
system. It must be emphasized, that only the production platform steady state response was
considered in this analysis.
Considering the Figure 12 (12a, 12c and 12e), related to the production platform steel deck
dynamic response, it can be verified the existence of several energy transfer areas with
several energy transfer picks, before the addition of the proposed bracing system.
The peak of interest for the forced vibration numerical analysis, associated with the
excitation frequency of the equipment (f = 30 Hz) is clearly indicated in this figure, see
Figure 12 (12a, 12c and 12e). However, after the introduction of the proposed bracing
system, the energy transfer pick related to the excitation frequency of the equipment was
significantly reduced, see Figure 12 (12b, 12d and 12f). This fact demonstrates the efficiency
of the proposed bracing system and indicates that the structural system dynamic response
can be significantly reduced.
Considering the same strategy previously presented, the design recommendations (see
Tables 10 to 12) were utilized, in order to determine the limiting values for machinery
performance, related to allowable amplitudes, velocities and accelerations (ISO 1940-1, 2003;
ISO 2631-1, 1997).

303

a) Node 9196 before the addition of the
bracing system
b) Node 9196 after the addition of the bracing
system

c) Node 9189 before the addition of the
bracing system
d) Node 9189 after the addition of the bracing
system

e) Node 9282 before the addition of the
bracing system
f) Node 9282 after the addition of the bracing
system

Fig. 12. Vertical displacements response spectra of the steel deck selected joints before and
after the introduction of the proposed bracing system
Mode: 1203
f = 30.0 Hz
Mode: 1203
f = 30.0 Hz
Mode: 1195
f = 30.0 Hz
Mode: 1195
f = 30.0 Hz
Mode: 1195
f = 30.0 Hz
Mode: 1203
f = 30.0 Hz


304

Rotor Support (Node:
9194) (m)
9197) (m)
9224) (m)
Amplitudes Limit
Value (m)
35 96 54
40 to 60*
Gear Support (Node:
9193) (m)
Gear Support (Node:
9196) (m)
Gear Support (Node:
9223) (m)
33 52 5
* For vertical vibration for high speed machines (>1500 rpm) (ISO 1940-1, 2003).

Table 13. Vertical displacements related to the combined dynamic loading (driving), after
the addition of the proposed bracing system

9194) (mm/s)
9197) (mm/s)
9224) (mm/s)
Velocities Limit
Value (mm/s)
6.68 6.92 10.36
0.70 to 2.80*
Gear Support (Node:
9193) (mm/s)
Gear Support (Node:
9196) (mm/s)
Gear Support (Node:
9223) (mm/s)
6.23 9.80 2.71
* Tolerable velocity for electrical motors according to (ISO 1940-1, 2003).

Table 14. Velocities related to the combined dynamic loading (driving), after the addition of
the proposed bracing system

Steel Deck
(Node: 9098)
(m/s
2
)
Steel Deck
(Node: 9137)
(m/s
2
)
Steel Deck
(Node: 9173)
(m/s
2
)
Steel Deck
(Node: 9189)
(m/s
2
)
Amplitudes Limit
Value (m/s
2
)
1.39 0.50 2.63 4.96
0.315 to 1.0*
Steel Deck
(Node: 9221)
(m/s
2
)
Steel Deck
(Node: 9282)
(m/s
2
)
Steel Deck
(Node: 9290)
(m/s
2
)
Steel Deck
(Node: 9317)
(m/s
2
)
3.44 0.20 1.23 2.75
* Acceptable acceleration values for human comfort in accordance with (ISO 2631-1, 1997).

Table 15. Accelerations related to the combined dynamic loading (driving), after the
addition of the proposed bracing system

305
This way, after the addition of the bracing system, see Figure 10, the structural system
dynamic response, submitted to the dynamic loadings (driving) coming from the machinery
presented the following results: the maximum amplitude value at the base of the driving
support (Node 9197: see Figure 9) on the platform steel deck was reduced of 446 m (or
0.446 mm or 0.0446 cm) to 96 m (or 0.096 mm or 0.0096 cm), see Tables 10 and 13; the
maximum velocity value calculated at the base of the driving support (Node 9197: see
Figure 9), on the steel deck was reduced of 84.12 mm/s to 6.92 mm/s, see Tables 11 and 14;
the peak acceleration value calculated on the structure steel deck was reduced of 27.33 m/s
to 4.96 m/s, see Tables 12 and 15.
It must be emphasized that despite the fact that the limiting values for machinery
performance, related to allowable amplitudes, velocities and accelerations (ISO 1940-1, 2003;
ISO 2631-1, 1997) were violated, even after the addition of the proposed bracing system, the
steady-state structural system dynamic response was considerably reduced.
8. Final remarks
This paper investigated the dynamic behaviour of an oil production platform made of steel
and located in Santos basin, So Paulo, Brazil. The structural model consists of two steel
decks with a total area of 1915 m
2
(upper deck: 445 m
2
and lower deck: 1470 m
2
), supported
by vertical sections made of tubular steel members (steel jacket), and piled into the seabed,
when subjected to impacts produced by mechanical equipment (rotating machinery). The
main objective of the paper was to assess the dynamic impacts on the steel deck structure
coming from the electrical generators and compressors.
Based on the peak acceleration values and maximum displacements and velocities, obtained
on the structure steady-state response, it was possible to evaluate the structural model
performance in terms of human comfort, maximum tolerances of the mechanical equipment
and vibration serviceability limit states of the structural system, based on the design code
recommendations.
It was concluded that the oil production platform displacements response spectra presented
several energy transfer areas with several energy transfer picks. However, after the
introduction of the horizontal steel bracing system proposed by the authors, the energy
transfer pick related to the excitation frequency of the equipment was significantly reduced,
demonstrating the efficiency of the idealized system.
After the addition of the horizontal bracing system the structural system dynamic response
presented the following results: the maximum amplitude value at the base of the driving
support on the steel deck was reduced of 0.0446 cm to 0.0096 cm; the maximum velocity
value calculated at the base of the driving support on the deck was reduced of 84.12 mm/s
to 6.92 mm/s and the maximum acceleration value calculated on the structure steel deck
was reduced of 27.33 m/s to 4.96 m/s.
However, even considering the improvement on the structural model bracing system and a
considerably reduction on the structure dynamic response, the results obtained throughout
this investigation indicated that the analysed platform steel deck violated the human
comfort criteria, as well as its vibration serviceability limit states, inducing that individuals
working temporarily near the machinery could be affected by human discomfort.


306
On the other hand, considering the machinery performance, it was also concluded that the
platform steel deck design, should be revaluated, due to the fact that the displacements and
velocities values related to the machinery supports were very high and violated the
recommended limits proposed by design codes.
9. Acknowledgements
The authors gratefully acknowledge the support for this work provided by the Brazilian
Science Foundations: CAPES, CNPq and FAPERJ.
10. References
Assuno, T. M. R. C. (2009). Consideraes sobre efeitos dinmicos e carregamentos induzidos
por fontes de excitao em estruturas. MSc Dissertation (In Portuguese), Civil
Engineering Post-graduate Programme, Federal University of Minas Gerais,
UFMG, Minas Gerais, MG, Brazil.
Bachmann, H.; Ammann, W. (1987). Vibrations in Structures: Induced by Man and Machines,
Structural Engineering Document 3
rd
Ed., International Association for Bridges and
Structural Engineering.
Chopra, A. K. (2007). Dynamics of Structures Theory and Applications to Earthquake
Engineering. 3
rd
Ed., Pearson Education, Inc., New Jersey.
Clough, R. W.; Penzien, J. (1995). Dynamics of Structures, 3
rd
Ed., Computers and Structures,
Inc., Berkeley.
Comit Euro-International Du Beton. (1991). Vibration Problems in Structures - Practical
Guidelines. Bullitin dnformation n 209.
Craig JR., R. R. (1981). Structural Dynamics An Introduction to Computer Methods. John
Wiley& Sons, Inc., New York.
Dias Junior, M. Dinmica de Rotores. Proceedings of IV Congresso da Academia
Trinacional de Cincias, Paran, PR, Brazil, October 2009, Fozdo
Iguau.
Figueiredo Ferraz Consultoria e Engenharia de Projeto LTDA. (2004). Anlise em Servio da
Plataforma de Merluza (PMLZ-1), Technical Report (In Portuguese).
Griffin, M. J. (1996). Handbook of Human Vibration. Ed. Academic Press, London.
GTSTRUDL. (2009). Structural Design & Analysis Software, Release 29.1.
ISO 1940-1. Mechanical Vibration. (2003). Balance Quality Requirements for Rotors in a Constant
(Rigid) State. Part 1: Specification and Verification of Balance Tolerances.
ISO 2631-1. Mechanical Vibration and Shock. (1997). Evaluation of Human Exposure to Whole-body
Vibration. Part 1: General Requirements.
ISO 2631-2. Mechanical Vibration and Shock. (1989). Evaluation of human exposure to whole-body
vibration. Part 2: human exposure to continuous and shock-induced vibrations in buildings
(1 to 80Hz).
Lenzen, K. H. (1996). Vibration of Steel Joist Concrete Slab Floors. In: Engineering journal,
v.3(3), p. 133-136.
Lpez, E. J. (2002). Dinmica de Rotores. Graduation Monography. Universidad Nacional del
Comahue.

307
Milet, R. R., (2006). Anlise Comparativa de Mtodos de Clculo para Fundaes de Mquinas,
MSc Dissertation (In Portuguese), Civil Engineering Post-graduate Programme,
Federal University of Pernambuco, UFPE, Pernambuco, PE, Brazil.
Murray, T. M., Allen, D. E. and Ungar, E. E. (2003). Floor vibration due to human activity, Steel
design guide series, AISC, Chicago, USA.
Pereira, C. C. G. (2005). Curvas de Percepo e Conforto Humano para Vibraes Verticais. MSc
Dissertation (In Portuguese), Civil Engineering Post-graduate Programme, Federal
University of Ouro Preto, UFOP, Minas Gerais, MG, Brazil.
Poulos, H. G.; Davis, E. H. (1980). Pile Foundation Analysis and Design. John Wiley & Sons
Inc., New York.
Reiher, H. E, Meister, F. J. (1946). The Effect of Vibration on people. Translated from Forsch
Geb, p 381-386, Ohio.
Rimola, B. D. (2010). Anlise Dinmica de Plataformas de Ao a partir da Considerao do Efeito
da Interao Solo-Estrutura, MSc Dissertation (In Portuguese), Civil Engineering
Post-graduate Programme, PGECIV, State University of Rio de Janeiro, UERJ, Rio
de Janeiro, RJ, Brazil.
Rimola, B. D. ; Silva, J. G. S. da ; Sieira, A. C. C. F. ; Lima, L. R. O. de ; Neves, L. F. da C.
(2010a). Vibration Analysis of a Production Platform Induced by Mechanical
Equipments. Proceedings of The Tenth International Conference on Computational
Structures Technology, CST 2010, Valncia, Spain. Civil-Comp Press, Edinburg.
v. 1. p. 1-12.
Rimola, B. D ; Silva, J. G. S. da ; Sieira, A. C. C. F. ; Lima, L. R. O. de. (2010b). Vibration
Analysis of an Oil Production Platform. Proceedings of the CILAMCE 2010: XXXI
Iberian Latin American on Computational Methods in Engineering and MECOM
2010: IX Argentinean Congress on Computational Mechanics, CILAMCE 2010,
Buenos Aires, Argentina. AMCA (Argentine Association for Computational
Mechanics). v. XXIX. p. 7529-7540.
Silva, E. L. da. (2004). Dinmica de Rotores: Modelo Matemtico de Mancais Hidrodinmicos.
MSc Dissertation (In Portuguese), Federal University Paran. Paran,PR,
Brasil.
Souza, M. G. de; Cicogna, T. R.; Chiquito, A. J. (2007). Excitao dos modos normais de um
sistema usando um motor desbalanceado. Revista Brasileira de Ensaio de Fsica, v. 29,
n. 1, p.5-10, Brazil.
Srinivasulu, P.; Vaidyanathan, C. V. (1976). Handbook of Machine Foundations. McGraw-Hill,
New Delhi.
Terzaghi, K. (1955). Evaluation of Coefficients of Subgrade Reaction. Gotechnique, London,
v. 4, n 4, p. 297-326.
Vasconcelos, R. P. (1998). Atenuaes de Vibraes em Lajes Nervuradas de Piso em Concreto
Armado. MSc Dissertation (In Portuguese), Civil Engineering Post-graduate
Programme, Federal University of Rio de Janeiro, UFRJ, Rio de Janeiro, RJ,
Brazil.
Winkler, E. (1867). Die Lehre von Elastizitat und Festigkeit (On Elasticity and Fixity),
Dominicus, Prague.


308
Zhou, S.; Shi, J. (2001). Active Balancing and Vibration Control of Rotating Machinery: A Survey.
The Shock and Vibration Digest - Sage Public ations, Vol. 33, No. 4,
p. 361-371.
0
MIMO Vibration Control for a Flexible Rail Car
Body: Design and Experimental Validation
Alexander Schirrer
1
, Martin Kozek
1
and Jrgen Schftner
2
1
Institute of Mechanics and Mechatronics, Vienna University of Technology
2
Institute of Technical Mechanics (ACCM), Johannes Kepler University of Linz
Austria
1. Introduction
Car bodies of modern rail vehicles are designed as lightweight structures with the aim
to minimize mass and thus operational energy demand. The central structural design
requirements are given by the main static and dynamic loads. However, ride comfort becomes
an increasingly important issue because the softer, more compliant structure exhibits low
eigenfrequencies that signicantly affect perceived passenger ride comfort.
Various approaches have been taken to reduce comfort-relevant vibrations of the car body
that can be grouped into vibration isolation and vibration damping approaches. The isolation
approaches include passive, semi-active, and active concepts to decouple the car body from
the bogeys and wheel sets. The active approaches are more complex and can affect the safety
against derailment, but potentially lead to improved isolation performance over their passive
counterparts (see Foo & Goodall (2000) and Stribersky et al. (1998)).
The complementary vibration damping approaches intend to increase the elastic eigenmodes
damping ratios. A passive approach is taken in Hansson et al. (2004), active control schemes
have been proposed in Kamada et al. (2005) as well as Schandl et al. (2007) and Benatzky
(2006). The latter two references treat the same metro conguration and actuation concept
as the present work.
This chapter presents LQG and weighted 1
2
MIMO control design methods for the vibration
control of lightweight rail car body structures. These designs are studied and compared to
achieve vibration reduction and passenger ride comfort improvement in a highly exible
metro rail car body. The metro car body structure is directly actuated via locally mounted
Piezo stack actuators. Utilizing strain measurement signals, the control law actuates the
structure with the aim of minimizing ride comfort-relevant acceleration signals across the car
body interior. This system is subject to variations in damping and frequency of the exible
modes which pose a challenge for control design; the two control methods are implemented
in a simulation as well as in a scaled experimental setup and their capabilities are investigated.
The work is embedded in a rich series of research and publications treating various aspects
of control design for exible rail car bodies. A robust 1
-optimal control approach

is surveyed in Kozek et al. (2011), including an overview on modeling, control design,
simulation, identication, and experimental results. These research topics are focused on and
detailed in Benatzky & Kozek (2005; 2007a;b); Benatzky et al. (2006; 2007); Bilik et al. (2006);
15
2 Vibration Control
Popprath et al. (2006; 2007) and Schirrer et al. (2008). Further studies focus on Piezo stack
actuators, their modeling and control (Schirrer, Kozek & Benatzky, 2008). Schirrer & Kozek
(2008) and Schirrer (2010) report co-simulation studies of the multi-body system dynamics,
the control law, and a non-linear actuator model. The underlying research project and its
results are documented in two doctorate theses (Benatzky, 2006; Schandl, 2005), as well as in
a series of master / diploma theses (Bilik, 2006; Schftner, 2006; Schirrer, 2010). This work is
based on the results of Schftner (2006), Benatzky (2006), and Schirrer & Kozek (2008).
This text should provide the readers with rst-hand experience of robust control design and
implementation. It is intended to relate control theory results, simulation, and experimental
results. This is done with the aim of improving ones understanding of relevant design
parameters, caveats, and ways to successfully establish a working control law for challenging
control problems. At the example of the application of modern control design methods to a
complex, three-dimensional vibration control problem, a reliable methodology for tackling
structurally new problems in vibration control is demonstrated. Additionally, the novel
application of large-scale Piezo stack actuators for the vibration control of large and heavy
exible structures is studied.
This chapter is structured as follows: In the rst section, a system description of the exible
metro car is given. The equations of motions are expressed by a rst-order state-space
representation obtained by a nite element analysis within the frequency range of interest.
Details about the actuation, measurement, and performance variables are introduced which
play an essential role for the MIMO controller design. Since the system order of the optimal
state-space controller is equal to the order of the open-loop system, system order reduction is
often necessary. The control laws designed for low-order plants then have to be validated
on the original system for closed-loop stability and performance. Two types of optimal
controllers are designed for a reduced-order plant model: an LQG and a weighted 1
2
controller. Simulation results are analysed and interpreted.
In an additional section, real-time control laws are designed with both design methodologies
for a laboratory scale model of a rail car body. The redesigned controllers are validated
in experimental test runs, providing an extended understanding on robustness, design
parameter tuning, and real-time implementation issues. Finally, conclusions are drawn from
both theoretical and experimental experience, and specic properties of the design methods
are discussed.
2. System description - modeling of a exible rail car body
2.1 Flexible rail car body model
For the different controller design procedures which will be explained in the next chapters a
mathematical model of the exible rail car body is necessary. The equations of motion of a
linear time-invariant system can be stated in state-space form as:
x = Ax +B
1
w+B
2
u
z = C
1
x +D
11
w+D
12
u (1)
y = C
2
x +D
21
w+D
22
u,
where the rst line describes the differential equation of the physical system. The second and
third lines are the output equations. The system matrix of the dynamic system and the input
MIMO Vibration Control for a Flexible Rail Car Body: Design and Experimental Validation 3
matrices for the disturbance wand the actuation u are denoted by A, B
1
, and B
2
, respectively.
The output matrices C
1
and C
2
as well as the feedthrough matrices D
11
, D
12
, D
21
, and D
22
depend on the locations and on the types of the measurements and the performance outputs.
In the literature on 1
2
and 1
controller design, z is called the output performance vector to

be optimized and y is the measurement vector.
The state-space model description (1) can be established through analytical modeling,
typically via the Finite Element (FE) Method, or through data-based identication. These steps
are detailed in Kozek et al. (2011) for the rail car body model. In this work, the simulation
part is based on an analytic model and the experimental part utilizes a model identied from
measured signal data.
Fig. 1 shows the rail car body containing the positions for the performance outputs (P
1
P
6
),
the disturbances (E
1
E
4
), the feedback sensors (S
1
S
4
), and the actuators (A
1
A
4
). For a
detailed discussion on optimal placement of sensors and actuators see Schirrer (2010).
E
1
S
1
P
1
S
2
P
2
E
2
P
3
S
P
5
P
6
... Actuator
... Sensor
... Performance
position
... Excitation
force
Symbols
A
4
A
2
Fig. 1. Flexible rail car body: positions of input and output variables
The mathematical model of the rail car body, obtained by an FE model and subsequent order
reduction steps, considers 17 elastic modes and 12 frequency-response-modes (FRM), see
Schandl (2005). It is thus of order 58.
Fig. 2 shows the minimum and maximum singular values of the open-loop transfer function
matrix T
dz
which may be interpreted as the best- and the worst-case functional relationship
from the white-noise disturbance vector d (related to w as shown below in Sec. 2.2) to the
performance vector z. From the singular values of the frequency response (which is identical
to the Bode magnitude response in the SISO case), one can observe the elastic modes in the
frequency range = 40 200
rad
/s (region 2, blue circles) as well as the FRM in the higher
frequency domain (region 3, red squares). Note that no suspension (rigid-body) modes are
modeled that typically lie at low frequencies (region 1), because only the exible car body
structure is considered. Full multi-body validation is studied via extensive co-simulations in
Schirrer (2010). The main goal in the next sections is to design controllers which suppress
elastic modes, but which are not active at other frequencies because the accuracy of the
simulation model decreases with increasing frequency. The transfer function is likely to be
dominated by other physical effects and nonlinearities for > 200
rad
/s which have not
been considered in obtaining the mathematical model. The FRM are special modes, which
311 MIMO Vibration Control for a Flexible Rail Car Body: Design and Experimental Validation
4 Vibration Control
are necessary to accurately describe highly localized deformations caused by the structure
actuation at the actuator interfaces (see Schandl (2005)). They lie at high frequencies and
can be treated here as testing modes, since the actuator action in the high-frequency domain is
undesired. High-frequency modes that are always present in continuous structures, but which
are not considered in the model used for the controller design may be excited and destabilize
the closed-loop system.
Frequency in
rad
/s
min./max. singular values T
dz
S
i
n
g
u
l
a
r
v
a
l
u
e
s
i
n
d
B
0
25
50
75
100
10
0
10
1
10
2
10
3
10
4
Region 1 Region 2
Region 3
open loop
modes 1 17
modes 18 29
Fig. 2. Minimum and maximum singular values of the open-loop system showing the elastic
modes. Modes denoted by squares are the frequency response modes (see Sec. 2.1).
2.2 Excitation
The rail vehicle runs on an imperfect track with misalignments and rail position variations.
This track excitation is often characterized by its spatial frequency spectrum and associated
lters, see Frederich (1984). For the studied rail car body dynamics, the disturbance wmodels
the excitation of the car body by secondary suspension forces. Its spectrum is known from
simulation and measurements and can be modeled by a colored noise sequence resulting from
ltering a white noise disturbance input signal d by a second-order lag function with cut-off
frequency
c.o.
= 10
rad
/s:
F(s) = 3000
100
s
2
+20s +100
. (2)
The colored noise w(t) is related to the white noise signal d(t) by
W(s) = F(s)D(s) w(t) = /
1
F(s)D(s) with d(t) = /
1
D(s). (3)
2.3 Actuators
A novel concept of structural actuation is applied here: Piezo stack actuators, mounted in
consoles at the oor support trusses of the car body structure locally introduce bending
moments into the structure. This allows direct actuation of the structural vibrations and is
thus considered efcient. In this study, four stack actuators are considered in the full-size
car body simulations (see Section 3 and Figure 1), whereas two stack actuators are utilized
in the experimental setup in Section 3.4, see Figure 25. For more detailed studies on the
actuator concept refer to Kozek et al. (2011); the deformation characteristics of a simple
console interface structure with a large-scale Piezo stack actuator have been investigated in
Schirrer et al. (2008).
console interface
car body
Piezo stack actuator
Fig. 3. Actuator mounted in a console to introduce torques into the attached exible structure
(rail car body)
2.4 Measurements
For the controller design, a signal-to-noise ratio (SNR) of about 10 is assumed, see Fig. 4.
Two types of measurements are available: either using the acceleration or using strain sensor
outputs proportional to the local structure curvature. The measurement positions are either
collocated with the actuator or located in the middle of the car body underneath the doors
(see Fig. 1). The latter case, when the actuator and sensor positions do not coincide, is called
non-collocation. In general, non-collocated systems suffer from a lack of robustness and
should not be used if the uncertainty of the system is large, but the controller performance
may be better than for collocated systems if a sufciently accurate mathematical model is
available, see Preumont (2006).
In the present work, four strain sensors respectively six acceleration sensors are utilized in the
full-size car body simulations. In the experimental setup, two strain sensors respectively one
acceleration sensor are utilized for feedback (however, ride comfort performance is evaluated
always at six acceleration performance positions as indicated in Figure 1).
Time in s
y
i
i
n
m
/
s
2
0.01
0.005
0
0.005
0.01
5.2 5.3 5.4
y
1
y
2
y
3
y
4
Fig. 4. Measured acceleration signals of the sensors showing a signal-to-noise ratio SNR = 10
6 Vibration Control
2.5 Evaluation of passenger ride comfort according to ISO 2631
Whole-body vibrations are transmitted to the human body of the passengers in a bus, train or
when driving a car. The ISO 2631 standard provides an average, empirically veried objective
quantication of the level of perceived discomfort due to vibrations for human passengers
(ISO, 1997). The accelerations in vertical and horizontal directions are ltered and these
signals root mean square (RMS) are combined into a scalar comfort quantity. Fig. 5 shows
the ISO 2631 lter magnitude for vertical accelerations which are considered the only relevant
component in the present study. For the heavy metro car, the highest sensitivity of a human
occurs in the frequency range of f 4 10 Hz. For the scaled laboratory model, all relevant
eigenfrequencies are shifted by a factor of 8 compared to the full-size FEM model. For this
reason, the ISO 2631 comfort lters and the excitation spectra are also shifted by this factor.
Moreover, only unidirectional vertical acceleration signals are utilized as they represent the
main contributions for the considered application.
Frequency in
rad
/s
ISO 2631-lter for rail vehicle ride comfort
shifted lter for laboratory model
M
a
g
n
i
t
u
d
e
i
n
d
B
25
0
25
50
10
0
10
1
10
2
10
3
10
4
Fig. 5. Filter function according to ISO 2631 (yaw axis)
3. Optimal controller design for the metro car body
Two different methods for controller design are investigated in the following: an LQG and a
frequency-weighted 1
2
controller are computed for a reduced-order plant model containing
only the rst 6 eigenmodes. The goal of this study is to obtain a deeper understanding on
robustness and controller parameter tuning, since the LQG and the frequency-weighted 1
2
control methods are applied to design real-time state-space controllers for the laboratory setup
in the next chapter.
3.1 LQG controller for a reduced-order system
3.1.1 Theory
The continuous-time linear-quadratic-gaussian (LQG) controller is a combination of an
optimal linear-quadratic state feedback regulator (LQR) and a Kalman-Bucy state observer,
see Skogestad & Postlethwaite (1996). Let a continuous-time linear-dynamic plant subject to
process and measurement noises be given in state space (D = 0 for compactness):
x = Ax +Bu +Ew
y = Cx +v, (4)
where w and v are assumed to be uncorrelated zero-mean Gaussian stochastic (white-noise)
processes with constant power spectral density matrices W and V.
The LQG control law that minimizes the scalar integral-quadratic cost function
J = E
_
lim
T
1
T
_
T
0
l(x, u)dt
_
(5)
with
l(x, u) = x
T
Qx +u
T
Ru (6)
turns out to be of the form
x = Ax +Bu + H(y Cx) (7)

u = K
LQR
x. (8)
Thereby, E[] is the expected value operator, Q = Q
T
_ 0 and R = R
T
~ 0 are constant,
positive (semi-)denite weighting matrices (design parameters) which affect the closed-loop
properties, (7) is the Kalman observer equation, and (8) is the LQR state feedback control law
utilizing the state estimate.
The optimal LQR state feedback control law (Skogestad & Postlethwaite, 1996)
u = K
LQR
x (9)
minimizes the deterministic cost function
J =
_

0
l(x, u)dt (10)
and is obtained by
K
LQR
= R
1
B
T
X, (11)
where X is the unique positive-semidenite solution of the algebraic Riccati equation
A
T
X +XAXBR
1
B
T
X +Q = 0. (12)
The unknown systemstates x can be estimated by a general state-space observer (Luenberger,
1964). The estimated states are denoted by x, and the state estimation error is dened by
:= x x. (13)
Choosing the linear relation
x = Fx +Gu + Hy, (14)

for state estimation, the following error dynamics is obtained:
= F + (A HC F)x + (B G)u. (15)
8 Vibration Control
If F = A HC and G = B hold, and if the real parts of the eigenvalues of F are negative, the
error dynamics is stable, x converges to the plant state vector x, and the observer equation (7)
is obtained.
With the given noise properties, the optimal observer is a Kalman-Bucy estimator that
minimizes E
_
(see Mohinder & Angus (2001); Skogestad & Postlethwaite (1996)). The
observer gain H in (7) is given by
H = YC
T
V
1
, (16)
where Y is the solution of the (lter) algebraic Riccati equation
AY +YA
T
YC
T
V
1
CY +EWE
T
= 0. (17)
Taking into account the separation principle (Skogestad & Postlethwaite, 1996), which states
that the closed-loop system eigenvalues are given by the state-feedback regulator dynamics
A BK together with those of the state-estimator dynamics A HC, one nds the stabilized
regulator-observer transfer function matrix
G
yu
(s) = K[sI A+ HC +BK]
1
H. (18)
Remark: The solutions to the algebraic Riccati equations (12) and (17) and thus the LQG
controller exist if the state-space systems
_
A, B, Q
1
2
_
and
_
A, W
1
2
, C
_
are stabilizable and
detectable (see Skogestad & Postlethwaite (1996)).
3.1.2 LQG controller design and results for strain sensors / non-collocation
The controller designs are based on a reduced-order plant model which considers only the
lowest 6 eigenmodes. The smallest and largest singular values of the system are shown in
Fig. 6 and Fig. 7 (compare Fig. 2 for the complete system). The eigenvalues are marked by
blue circles. The red lines depict the singular values of the order-reduced T
dz,red
(including
the shaping lter (2) for the colored noise of the disturbance signal w).
Since a reduced-order system is considered for the controller design, the separation principle
is not valid any longer for the full closed-loop system. Neither the regulator gain K
LQR
nor
the estimator gain H is allowed to become too large, otherwise spillover phenomena may
occur that potentially destabilize the high-frequency modes. Therefore, the design procedure
is an (iterative) trial-and-error loop as follows: in a rst step, the weighting matrices for the
regulator are prescribed and the resulting regulator gain is used for the full-order system
where it is assumed that the state vector can be completely measured. If spillover occurs, the
controller action must be reduced by decreasing the state weighting Q. In a second step, the
design parameters for the Kalman-Bucy-lter are chosen, considering the fact that the process
noise w is no white noise sequence any longer, see (2). Since the process noise covariance is
approximately known as (84.54 N)
2
for each channel, the weighting for the output noise V is
utilized as a design parameter.
For the optimal regulator the weighting matrices for the states and the input variables are
chosen as
Q = 9 10
8
I
1212
, R = I
44
, (19)
0
Frequency in
rad
/s
S
i
n
g
u
l
a
r
v
a
l
u
e
s
i
n
d
B
wz,red
and T
dz,red
T
wz,red
open loop
T
dz,red
open loop with comfort lter
25
25
50
75
100
10
0
10
1
10
2
10
3
10
4
Fig. 6. Smallest and largest singular values of the reduced-order open-loop system (6 modes)
Frequency in
rad
/s
wz,red
S
i
n
g
u
l
a
r
v
a
l
u
e
s
i
n
d
B
30
30
10
10
30
60 100 200
Fig. 7. Smallest and largest singular values of the reduced-order open-loop system (6 modes,
zoomed)
where I
nn
is the identity matrix (n rows, n columns). The observer weightings are chosen to
be
W = 84.54
2
I
44
, V = (1.54 10
6
)
2
I
44
. (20)
Table 1 lists the reduction of the ISO-ltered (see Fig. 5) RMS of each performance variable
z
1,ISO
z
6,ISO
compared to open-loop results. Figures 811 contain the maximum/minimum
singular values from the white noise input d (which is related to the colored noise input w
by (3)) to the performance vector z, the time-domain response of two selected performance
variables z
1
and z
6
, and two pole location plots (overview and zoomed) for the open- and the
closed-loop results.
Performance position index i 1 2 3 4 5 6 avg.
RMS reduction z
i,ISO
in % 8.44 11.22 29.64 26.53 30.05 31.80 22.94
Table 1. RMS reduction of the performance vector z by LQG control (strain sensors /
non-collocation), system order 12
open loop
closed loop
max./min. singular values T
dz
Frequency in
rad
/s
S
i
n
g
u
l
a
r
v
a
l
u
e
s
i
n
d
B
0
10
20
30
40
50
60
30 100 300
Fig. 8. Reduction of rail car disturbance transfer singular values with non-collocated LQG
control
open loop
closed loop
z
1
z
6
0
0
0.01
0.01
0.01
0.01
6
6
6.25
6.25
6.5
6.5
6.75
6.75
7
7
Time in s
Fig. 9. Acceleration signals z
1
and z
6
without/with non-collocated LQG control
open loop
closed loop
Re
I
m 0
0
2000
4000
2000
4000
20 40 60 80
Fig. 10. Rail car model open-loop and non-collocated LQG closed-loop pole locations
open loop
closed loop
Re
I
m0
0
200
100
100
200
1 2 3 4
Fig. 11. Rail car model open-loop and non-collocated LQG closed-loop pole locations
(zoomed)
3.1.3 Controller design and results for acceleration sensors / collocation
The optimal regulator is designed with the same weighting matrices for the states and the
control variables as for the case strain sensors / non-collocation, see (19). The observer
weightings are chosen to be
W = 84.54
2
I
44
, V = 0.154
2
I
44
. (21)
z
1,ISO
z
6,ISO
variables z
1
and z
6
RMS reduction z
i,ISO
in % 7.83 8.36 8.04 7.02 8.79 10.23 8.38
Table 2. RMS reduction of the performance vector z by LQG control (acceleration sensors /
collocation), system order 12

open loop
closed loop
dz
Frequency in
rad
/s
S
i
n
g
u
l
a
r
v
a
l
u
e
s
i
n
d
B
0
10
20
30
40
50
60
30 100 300
Fig. 12. Reduction of rail car disturbance transfer singular values with collocated LQG
control

open loop
closed loop
z
1
z
6
0
0
0.01
0.01
0.01
0.01
6
6
6.25
6.25
6.5
6.5
6.75
6.75
7
7
Time in s
1
and z
6
without/with collocated LQG control
open loop
closed loop
Re
I
m 0
0
2000
4000
2000
4000
20 40 60 80
Fig. 14. Rail car model open-loop and collocated LQG closed-loop pole locations
open loop
closed loop
Re
I
m 0
0
200
100
100
200
1 2 3 4
Fig. 15. Rail car model open-loop and collocated LQG closed-loop pole locations (zoomed)
3.2 Frequency-weighted 1
2
controller for a reduced-order system
The LQG controllers designed in the previous section do not take into account the
performance vector z. The design of the regulator and the estimator gains are a trade-off
between highly-damped modes, expressed by the negative real part of the closed-loop poles,
and robustness considerations. The generalization of the LQG controller is the 1
2
controller,
which explicitly considers the performance vector (e.g. one can minimize the deection
2-norm at a certain point of a exible system). Another advantage of this type of optimal
controller is the possibility to utilize frequency-domain weighting functions. In doing so, the
controller action can be shaped for specic target frequency ranges. In turn, the controller
can be designed not to inuence the dynamic behaviour where the mathematical model is
uncertain or sensitive to parameter variations.
high-pass lter low-pass lter
P
(s)
P(s)
K(s)
W
act
(s)
W
perf
(s)
u
w
z
y
Fig. 16. Closed-loop systemP(s) with controller K(s) and actuator and performance
weighting functions W
act
(s) and W
perf
(s)
Fig. 16 shows the closed-loop system, where the system dynamics, the controller, and the
frequency-weighted transfer functions are denoted by P(s), K(s), W
act
(s), and W
perf
(s).
Taking into account the frequency-weights in the system dynamics, the weighted system
description of P
can be formulated:
z
y
=
_
P
11
(s) P
12
(s)
P
21
(s) P
22
(s)
_
w
u
, (22)
where P
11
(s), P
12
(s), P
21
(s), and P
22
(s) are the Laplace domain transfer functions from the
input variables u and w to the output variables y and z.
3.2.1 1
2
control theory
Let the system dynamics be given in the state-space form (1), fullling the following
prerequisites (see Skogestad & Postlethwaite (1996)):
(A, B
2
) is stabilizable
(C
2
, A) is detectable
D
11
= 0, D
22
= 0
D
12
has full rank
D
21
has full rank
AjI B
2
C
1
D
12
has full column rank for all

AjI B
1
C
2
D
21
has full row rank for all

For compactness the following abbreviations are introduced:
R = D
12
T
D
12
S = B
2
R
1
B
2
T
A = AB
2
R
1
D
12
T
C
1
Q = C
1
T
C
1
C
1
T
D
12
R
1
D
12
T
C
1
_ 0
R = D
21
D
21
T
S = C
2
T
R
1
C
2
A = AB
1
D
21
T
R
1
C
2
Q = B
1
B
1
T
B
1
D
21
T
R
1
D
21
B
1
T
_ 0,
where _ 0 denotes positive-semideniteness of the left-hand side. The 1
2
control design
generates the controller transfer function K(s) which minimizes the 1
2
norm of the transfer
function T
wz
, or equivalently
|T
wz
|
2
=
1
2
_

T
wz
T
(j)T
wz
(j)d min. (23)
The controller gain K
c
and the estimator gain K
f
are determined by
K
c
= R
1
(B
2
T
X
2
+D
12
T
C
1
) (24)
and
K
f
= (Y
2
C
2
T
+B
1
D
21
T
)R
1
, (25)
where X
2
_ 0 and Y
2
_ 0 are the solutions of the two algebraic Riccati equations
X
2
A+ A
T
X
2
X
2
SX
2
+Q = 0, (26)
AY
2
+Y
2
A
T
Y
2
SY
2
+Q = 0. (27)
The state-space representation of the controller dynamics is given by
x = (A B
2
K
c
K
f
(C
2
D
22
K
c
))x +K
f
y
u = K
c
x,
u = K(s)y. (28)
3.2.2 1
2
controller design and results for strain sensors / non-collocation
The frequency-weighting functions have been specied as
W
act
= G
act
I
44
= 4967
(s +45)
4
(s
2
+6s +3034)
(s +620)
4
(s +2000)
2
I
44
(29)
W
perf
= G
perf
I
66
= 20 I
66
(30)
As in the previous section, the 1
2
controller is designed for the reduced-order model
(12 states). Considering the shaping lter (2) for the disturbance (8 = 4 2 states) and the
weighting functions (29) and (30) (24 = 4 6 states), one nds a controller of order 44.
z
1,ISO
z
6,ISO
variables z
1
and z
6
RMS reduction z
i,ISO
in % 26.27 27.95 28.71 27.84 30.99 34.31 29.35
Table 3. RMS reduction of the performance vector z by 1
2
control (strain sensors /
non-collocation), system order 44
open loop
closed loop
dz
Frequency in
rad
/s
S
i
n
g
u
l
a
r
v
a
l
u
e
s
i
n
d
B
0
10
20
30
40
50
60
30 100 300
Fig. 17. Reduction of rail car disturbance transfer singular values with non-collocated 1
2
control
3.2.3 1
2
controller design and results for acceleration sensors / collocation
The frequency-weighting functions have been specied as
W
act
= G
act
I
44
= 4967
(s +45)
4
(s
2
+6s +3034)
(s +620)
4
(s +2000)
2
I
44
, (31)
W
perf
= G
perf
I
66
= 20 I
66
. (32)
z
1,ISO
z
6,ISO

open loop
closed loop
z
1
z
6
0
0
0.01
0.01
0.01
0.01
6
6
6.25
6.25
6.5
6.5
6.75
6.75
7
7
Time in s
1
and z
6
without/with non-collocated 1
2
control
open loop
closed loop
Re
I
m0
0
2000
4000
2000
4000
20 40 60 80
Fig. 19. Rail car model open-loop and non-collocated 1
2
closed-loop pole locations
variables z
1
and z
6
RMS reduction z
i,ISO
in % 23.89 28.12 27.23 24.67 28.85 31.27 27.34
Table 4. RMS reduction of the performance vector z by 1
2
control (acceleration sensors /
collocation), system order 44
3.3 Interpretation
The main goal for both the LQG and the 1
2
controller designs was to increase the damping
of the rst three eigenmodes. In the present design task, the LQG controller designed for
open loop
closed loop
Re
I
m 0
0
200
100
100
200
1 2 3 4
Fig. 20. Rail car model open-loop and non-collocated 1
2
closed-loop pole locations (zoomed)
open loop
closed loop
dz
Frequency in
rad
/s
S
i
n
g
u
l
a
r
v
a
l
u
e
s
i
n
d
B
0
10
20
30
40
50
60
30 100 300
Fig. 21. Reduction of rail car disturbance transfer singular values with collocated 1
2
control
collocated acceleration sensors (see Section 3.1.3) did not yield satisfactory performance.
The singular value plot shows only marginal magnitude reduction (Figure 12), and also
a time-domain analysis of the performance signals z
1
and z
6
(see Figure 13) shows no
signicant improvement. According to Table 2, the reduction of the ltered performance
vector is approximately 8%. However, at 1500
rad
/s one of the frequency response modes
approaches the imaginary axis (Fig. 14 and Fig. 15). Even though the simulated closed loop
remains stable, this spillover is critical for operation at an uncertain real plant which possesses
unknown high-frequency dynamics.
Considering the LQG design for non-collocated strain sensors in Section 3.1.2, the controller
signicantly improves the vibrational behaviour. The performance vector is reduced by 23%
(Table 1) and a signicant reduction is apparent for the time-domain evaluation in Fig. 9. The
maximum singular values of the rst three eigenmodes are reduced (e.g. third eigenmode
open loop
closed loop
z
1
z
6
0
0
0.01
0.01
0.01
0.01
6
6
6.25
6.25
6.5
6.5
6.75
6.75
7
7
Time in s
1
and z
6
without/with collocated 1
2
control
open loop
closed loop
Re
I
m0
0
2000
4000
2000
4000
20 40 60 80
Fig. 23. Rail car model open-loop and collocated 1
2
closed-loop pole locations
11 dB, see Fig. 8). From the pole location plot one concludes that in the higher frequency
domain the frequency response modes remain unchanged (Fig. 10 and Fig. 11).
Both variants of the 1
2
-optimal controllers (Section 3.2.2 and Section 3.2.3) show signicantly
higher performance in simulation than the controllers obtained by the LQGdesign procedure.
The main advantage of the 1
2
design approach is the possibility to directly incorporate
frequency weights to shape the design, see (29) and (31). Specically, the frequency
content of the actuator command signals can be modied. The control law actuates
mainly within the frequency range 50 70
rad
/s due to the transmission zeros in the
weighting functions W
act
. In the high-frequency domain, W
act
is large for both 1
2
designs,
so only small actuator signal magnitudes result at these frequencies which is especially
open loop
closed loop
Re
I
m0
0
200
100
100
200
1 2 3 4
Fig. 24. Rail car model open-loop and collocated 1
2
closed-loop pole locations (zoomed)
favorable if the plant dynamics is unknown there. The results for strain/non-collocation and
acceleration/collocation control designs are shown in Figs. 1720 and Table 3 as well as in
Figs. 2124 and Table 4. In the rst case the vibrations of the time-domain performance signals
z
i,ISO
are reduced by 30% (Fig. 18), which is also indicated by the singular values plot (Fig. 17):
the lowest three modes are reduced on average by 11 dB. Virtually no spillover occurs at high
frequencies ( 150 4000
rad
/s): The singular values are unchanged (not shown) and also
the pole locations remain unchanged for > 150
rad
/s (seen in Fig. 19 and Fig. 20 where the
open-loop poles (blue circles) and closed-loop poles (black crosses) coincide).
The acceleration sensor / collocation simulation results show similar improvement: Only the
rst three modes are strongly damped (Fig. 21 and Fig. 24), the other ones are hardly affected
by the controller action due to the specic choice of the weighting function (31), see Fig. 23.
The average reduction of the ISO-ltered performance variables is 27% (Table 4 and Fig. 22).
As a concluding remark, note that the combination of the 1
2
method with
frequency-weighted transfer functions for the input and the performance signals (W
act
,
W
perf
) provide satisfactory results, which are characterized by their high robustness and
insensitivity to parameter uncertainties. It is shown that the frequency content of the
controller action can be tuned by the input weight W
act
, which affects only the rst modes
of interest. Higher modes, which are much more difcult to model, are hardly affected due
to the roll-off of the 1
2
controller. Nevertheless, the LQG controller shows very promising
results for the case of non-collocated strain sensors, although the controller is designed for
a strongly reduced model containing only 6 modes (note that the full order model has 29
modes). If the acceleration signals are measured and sensor and actuators are collocated, the
full-order plant is destabilized by the LQG controller (designed on the reduced-order plant).
Finally, it is noted that so-called reduced-order LQG controllers (see Gawronski (2004)) also
have been designed to control the metro vehicle, see Schftner (2006). By this method an
LQG controller has been directly designed for the full-order plant model with 29 modes.
Then, the controller transfer functions are evaluated (dynamic systems of order 58) and
transformed to the Gramian-based input/output-balanced form. Hardly observable or
controllable states, indicated by small Hankel singular values, are truncated, yielding a
low-order controller. While this procedure works well for academic problems (for example, a
simply-supported beam), for the metro car body no low-order controller with good vibration
reduction performance could be found, see Schftner (2006).
3.4 Experimental setup of scaled metro car body
3.4.1 General remarks
Fig. 25 and Fig. 26 show the laboratory testbed in which the metro car body scale model is
operated. The aluminum structure is excited via an electrodynamic shaker, two Piezo patches
measure local structure strain, and two Piezo stack actuators, mounted in consoles on the
structure, provide an efcient means of structural actuation. Fig. 26 also shows the actuation
and measurement setup symbolically with actuator amplier (AA), shaker amplier (SA),
antialiasing lters (AF), measurement amplier (MA) and the laboratory computer (Lab.-PC)
on which the real-time control algorithms are implemented.
Fig. 25. Scaled metro car body
Fig. 26. Basic sketch of the scaled metro car body with actuators, sensors, and performance
variables
The pole plot and the singular-value plot (Fig. 27 and Fig. 28) of the frequency response
provide information on the identied dynamics of the laboratory setup (200 modes): the
modes relevant for the control problem are the bending mode at f 65 Hz and the torsional
mode at f 75 Hz. The majority of the poles are either negligible high-frequency modes or
other local oscillatory modes.

Re
I
m
torsion
mode
bending
mode
suspension
mode
0
0
1000
500
500
1000
10 8 6 4 2 2
Fig. 27. Pole plot of the identied scaled metro car body (zoomed)
Singular values T
dz
S
i
n
g
u
l
a
r
v
a
l
u
e
s
i
n
d
B
torsion mode
bending mode
suspension
mode
Frequency in
rad
/s
0
30
20
10
10
20
100 500 1000
Fig. 28. Singular values of the identied scaled metro car body
The goal is to signicantly dampen the torsional and the bending modes without destabilizing
other oscillatory modes. For an objective evaluation of the active vibration control problem,
the RMS of the frequency-ltered performance variables z
i,ISO
(i = 1, . . . , 6) are compared
in the open-loop and closed-loop responses. These six performance quantities represent
a quantication of passenger ride comfort. A more detailed analysis can be found in
Kozek & Benatzky (2008) and Schirrer (2010).
3.4.2 LQG controller design
An LQG controller is designed for a reduced-order plant model with 28 states (the system
is identied with 200 states, see Kozek et al. (2011)). Hence the system is transformed into
a modal state space representation and all eigenmodes with an eigenfrequency higher than
f 75 Hz are truncated and not considered for the controller design. The weighting functions
for the regulator and the estimator design are
Q =
_
I
2424
0
0 6 10
6
I
44
_
, R = I
22
(33)
W = 0.0054, V = 2 10
4
I
22
. (34)
Note that Q in (33) is chosen such that only the bending and the torsional vibrations should
be signicantly damped. A discussion of the results for the closed-loop system is given in
Section 3.4.4.
3.4.3 1
2
controller design
The 1
2
controller is designed for a plant model which only considers 3 eigenmodes (two
of them describe the bending and torsional behaviour). For an efcient control design the
frequency-dependent actuator and performance functions are specied as
W
act
= G
act
I
22
= 188.5 10
9
(s
2
+77.91s +151800)
4
(s +10000)
8
I
22
. (35)
W
perf
= G
perf
I
66
= I
66
(36)
Note that the transmission zeros of W
act
are near the two target modes to be damped, causing
the actuator action to be a maximum for these frequencies. A discussion of the results for the
closed-loop system are given in Section 3.4.4.
3.4.4 Results
Table 5 and Figures 29, 30, 31, and 32 show the damping ability of both types of controllers. In
both cases, the vibrations of the actively controlled systemare signicantly reduced compared
to the open-loop response of the system. The accelerations at both ends of the structures,
expressed by the performance variables z
1
, z
2
, z
5
and z
6
can be signicantly reduced, whereas
z
3
and z
4
are close to the open-loop response. This is explained due to the fact that the
rst torsional mode dominates the bending vibrations for the scale laboratory setup. For the
LQG controller the singular values only differ for the torsional and the bending vibrations
(14 dB and 10 dB). It is evident that only the eigenvalues of the two targeted exible
modes are affected by the controller. The unchanged mode at lower frequency is an almost
uncontrollable suspension mode, while the higher exible modes are not adversely affected
by the control action.
Analogous results could be obtained using the frequency-weighed 1
2
-optimal control design
methodology: the achieved RMS reductions of the performance variables are approximately
the same as for the LQG control method. Note that the bandwidth of the frequency-weighted
controller is narrowaround f = 60 75 Hz where the actuator weightings are small. Contrary
to the LQGapproach, the target modes as well as other modes with a higher negative real part
are positively inuenced. This indicates that the model quality is sufciently high and that
the control laws are insensitive to the occurring differences between design plant and actual
system.
RMS reduction z
i,ISO
(LQG) in % 41.53 34.69 8.83 6.02 37.28 36.94 27.55
RMS reduction z
i,ISO
(1
2
) in % 41.31 35.69 -4.00 4.59 34.89 36.16 24.77
Table 5. Laboratory testbed results: RMS reduction of the performance vector z by an LQG
(system order 28) and an 1
2
controller (system order 23) utilizing strain feedback sensors

open loop
closed loop
Re
I
m0
0
1000
500
500
1000
80 60 40 20 20
Fig. 29. Pole plot with/without LQG controller
0
10
20
30

open loop
closed loop
Singular values T
dz
S
i
n
g
u
l
a
r
v
a
l
u
e
s
i
n
d
B
Frequency in
rad
/s
10
20
100 500 1000
Fig. 30. Singular values of the frequency response plot with/without LQG controller

open loop
closed loop
Re
I
m
bandwidth
0
0
1000
500
500
1000
80 60 40 20 20
Fig. 31. Pole plot with/without 1
2
controller

open loop
closed loop
Singular values in dB
S
i
n
g
u
l
a
r
v
a
l
u
e
s
i
n
d
B
Frequency in
rad
/s
0
30
20
10
10
20
100 500 1000
Fig. 32. Singular values of the frequency response plot with/without 1
2
controller
4. Conclusions
This chapter presents a case study on the design of MIMO control laws to reduce vibrations
in a exible metro rail car body and thus to improve passenger ride comfort. Direct structural
actuation by Piezo actuators is considered and two sensor concepts strain sensors (in a
non-collocated setting) and acceleration sensors (collocated) are evaluated. One part of
the outlined studies focused on a simulation model of a full-size lightweight metro rail car
body; the other part tests the control concepts on a laboratory testbed with a scale model
of the car body. The control laws have been designed by LQG and by frequency-weighted
1
2
-optimal control design methodologies. Both design methods are rst studied in the
simulation and compared. It is found that the weighted 1
2
designs yield controllers that
perform satisfactorily in the presence of model uncertainty and independent of the sensor
concept (strain sensors / non-collocation or acceleration sensors / collocation): the rst three
modes of interest are signicantly attenuated and the unknown modes in the high-frequency
domain are hardly affected by the controller action, thus increasing the ride comfort for
the passengers. The LQG controller minimizes the vibrations only for strain sensors in
the non-collocated setup. Finally, both design methods, which have been studied for the
lightweight rail car body simulations, are successfully implemented in a scaled laboratory
setup: it is demonstrated that the target modes (torsion and bending) have been signicantly
damped by both controller types. A further advantage of the weighted 1
2
controller is that
the controller action can be tuned for a specic bandwidth in the frequency domain, which is
essential if the dynamics of the structure under consideration is uncertain or a control input is
not desired for certain frequencies. The studies results show the applicability of weighted 1
2
control for partially uncertain exible-structure systems. The control goal of improving ride
comfort is directly formulated as a weighted 1
2
minimization problem which justies the
presented study. However, a range of related publications show the design and application
of robust 1
-optimal controllers for this application, which can give robustness guarantees
based on the structured singular value.
5. References
Benatzky, C. (2006). Theoretical and experimental investigation of an active vibration damping
concept for metro vehicles, PhD thesis, Institute for Mechanics and Mechatronics,
Division of Control and Process Automation, Vienna University of Technology,
Austria.
Benatzky, C. & Kozek, M. (2005). Effects of local actuator action on the control of large exible
structures, Proceedings of the 16th IFAC World Congress, Prague, Chech Republic.
Benatzky, C. & Kozek, M. (2007a). An actuator fault detection concept for active vibration
control of a heavy metro vehicle, Proceedings of the 14th International Congress on Sound
and Vibration (ICSV14), Cairns, Australia.
Benatzky, C. & Kozek, M. (2007b). An identication procedure for a scaled metro vehicle -
exible structure experiment, Proceedings of the European Control Conference ECC 2007,
Kos, Greece, Kos, Greece.
Benatzky, C., Kozek, M. & Bilik, C. (2006). Experimental control of a exible beam using a
stack-bending-actuator principle, Proceedings of the 20th Scientic Conference, Hanoi,
Vietnam.
Benatzky, C., Kozek, M. & Jrgl, H. (2007). Comparison of controller design methods for a
scaled metro vehicle - exible structure experiment, Proceedings of the 26th American
Control Conference, New York, USA.
Bilik, C. (2006). Aufbau und Inbetriebnahme des Prfstandmodelles eines
Schienenfahrzeug-Wagenkastens zum Nachweis von aktiver Schwingungsdmpfung,
Diploma thesis, Vienna University of Technology, Vienna.
Bilik, C., Benatzky, C. &Kozek, M. (2006). A PC-based multipurpose test bed environment for
structural testing and control, Proceedings of the 3rd International Symposium on Remote
Engineering and Virtual Instrumentation, Maribor, Slovenia.
Foo, E. & Goodall, R. M. (2000). Active suspension control of exible-bodied railway vehicles
using electro-hydraulic and electro-magnetic actuators, Control Engineering Practice
8(5): 507518.
Frederich, F. (1984). Die Gleislage - aus fahrzeugtechnischer Sicht, Vol. 108 (12) of Gleislauftechnik,
Siemens Verlagsbuchhandlung, pp. 355 361.
Gawronski, W. (2004). Advanced structural dynamics and active control of structures, Springer,
New York.
Hansson, J., Takano, M., Takigami, T., Tomioka, T. & Suzuki, Y. (2004). Vibration Suppression
of Railway Car Body with Piezoelectric Elements, JSME International Journal Series C
47(2): 451456.
ISO (1997). ISO2631-1: Mechanical vibration and shock - evaluation of human exposure to
whole-body vibration. Part 1: General requirements, International Organization for
Standardization. Corrected and reprinted July 15th, 2007.
Kamada, T., Tohtake, T., Aiba, T. & Nagai, M. (2005). Active vibration control of the railway
vehicle by smart structure concept, in S. Bruni & G. Mastinu (eds), 19th IAVSD
Symposium - Poster Papers.
Kozek, M. & Benatzky, C. (2008). Ein mastbliches Experiment zur aktiven
Schwingungsdmpfung eines Eisenbahn-Wagenkastens, at - Automatisierungstechnik
10(56): 504512.
Kozek, M., Benatzky, C., Schirrer, A. & Stribersky, A. (2011). Vibration damping of a exible
car body structure using piezo-stack actuators, Control Engineering Practice 19(3): 298
310. Special Section: IFAC World Congress Application Paper Prize Papers.
URL: http://www.sciencedirect.com/science/article/B6V2H-4X3MR4Y-2/2/3ef1d868e70c2b
6f10fd9412f9c8c1de
Luenberger, D. G. (1964). Observing the state of a linear system, IEEE Transactions on Military
Electronics 8(2): 7480.
Mohinder, S. & Angus, P. (2001). Kalman Filtering: Theory and Practice Using MATLAB, Wiley
Interscience, John Wiley & Sons, USA.
Popprath, S., Benatzky, C., Bilik, C., Kozek, M., Stribersky, A. & Wassermann, J. (2006).
Experimental modal analysis of a scaled car body for metro vehicles, Proceedings of
the 13th International Congress on Sound and Vibration (ICSV13), Vienna, Austria.
Popprath, S., Schirrer, A., Benatzky, C., Kozek, M. & Wassermann, J. (2007). Experimental
modal analysis of an actively controlled scaled metro vehicle car body, Proceedings of
the 14th International Congress on Sound and Vibration (ICSV14), Cairns, Australia.
Preumont, A. (2006). Mechatronics: Dynamics of Electromechanical and Piezoelectric Systems,
Springer.
Schandl, G. (2005). Methodenuntersuchung zur aktiven Schwingungsreduktion eines
Schienenfahrzeugwagenkastens, PhD thesis, Vienna University of Technology, Vienna.
Schandl, G., Lugner, P., Benatzky, C., Kozek, M. & Stribersky, A. (2007). Comfort enhancement
by an active vibration reduction systemfor a exible railway car body, Vehicle System
Dynamics 45(9): 835847.
Schftner, J. (2006). Aktive Schwingungsdmpfung eines Schienenfahrzeugwagenkastens durch
1
2
-Regelung, Masters thesis, Institute for Mechanics and Mechatronics, Division of
Control and Process Automation, Vienna University of Technology.
Schirrer, A. (2010). Co-Simulation of Rail Car Body Vibration Control with SimPACK
, VDM
Verlag Dr. Mller, Saarbrcken, Germany.
Schirrer, A. & Kozek, M. (2008). Co-simulation as effective method for exible structure
vibration control design validation and optimization, Control and Automation, 2008
16th Mediterranean Conference on, pp. 481 486.
Schirrer, A., Kozek, M. & Benatzky, C. (2008). Piezo stack actuators in exible structures:
Experimental verication of a nonlinear modeling and identication approach, 6th
EUROMECH Nonlinear Dynamics Conference (ENOC 2008), OPEN-ACCESS library.
URL: http://lib.physcon.ru
Schirrer, A., Kozek, M., Plank, A., Neumann, M., Badshah, S. & Wassermann, J.
(2008). Vibration analysis of an actively controlled exible structure using speckle
interferometry, Proceedings of 15th International Congress on Sound and Vibration
(ICSV15).
Skogestad, S. & Postlethwaite, I. (1996). Multivariable feedback control, John Wiley & Sons.
Stribersky, A., Mller, H. & Rath, B. (1998). The development of an integrated suspension
control technology for passenger trains, Proceedings of the Institution of Mechanical
Engineers, Part F: Journal of Rail and Rapid Transit, Vol. 212, pp. 3342.
16
Changes in Brain Blood Flow on Frontal Cortex
Depending on Facial Vibrotactile Stimuli
Hisao Hiraba
1
, Takako Sato
2
, Satoshi Nishimura
2
, Masaru Yamaoka
3
,
Motoharu Inoue
1
, Mitsuyasu Sato
1
, Takatoshi Iida
1
, Satoko Wada
1
,
Tadao Fujiwara
3
and Koichiro Ueda
1

1
Departments of Dysphasia Rehabilitation

2
Oral and Maxillofacial Surgery

and
3
Physics, Nihon University, School of Dentistry
Japan
1. Introduction
We provide patients who have problems with reduced salivation (hyposalivation) with
artificial saliva treatment, humectants, and salivary gland massage (Ueda et al. 2005).
However, treatment with artificial saliva and humectants is symptomatic, and although
salivary gland massage can reinvigorate weak glands, to do so is difficult for people with
disabilities and has varying effects, depending on operator skill. Thus, we have focused on
increasing salivation through the use of vibrotactile stimulation, as reported by Hiraba et al.
(2008). Before using this apparatus on patients, it was necessary to first estimate the effect on
normal subjects.
The biggest challenge with continuous use of stimulation is an adaptive effect. In particular,
we were interested in determining whether the effect was continuous without attenuation,
when patients continue using the apparatus every day (Despopoulos and Silbernagel, 2003).
We investigated adaptation to the continuous use of vibrotactile stimuli for 4 or 5 days in
the same subjects to determine whether this resulted in a decrease in salivation
(Despopoulos and Silbernagel, 2003; Principles of Neural Science. 2000a). Before this
experiment was performed, it was necessary to compare resting and stimulating salivary
secretion and to investigate the most effective frequency for increasing salivary secretion.
We examined the amount of salivation during vibrotactile stimuli with a single motor (1.9
m amplitude) on the bilateral masseter muscle belly (on the parotid glands), using a dental
cotton roll positioned at the opening of the secretory duct for 3 min. Furthermore, we
examined the amount of salivation during vibrotactile stimuli with single and double
motors (1.9 m and 3.5 m amplitudes) on the bilateral submandibular angles (on the
submandibular glands). Then, we compared resting and stimulating salivation and
investigated the most effective frequency for increasing salivary secretion. The effect of
increased salivation in normal subjects was determined as the difference between resting
and stimulating salivation.
We defined a 5-min interval as the recovery time between resting and stimulating salivation
from a preliminary study. First, we examined the most effective frequency for salivation of

the parotid glands among 89, 114, and 180 Hz with a single motor, and then we found the
most effective frequency for salivation of the submandibular glands between 89 and 114 Hz
with single and double motors. We discuss the effects of vibrotactile stimulation based on
these results.
Furthermore, to study the mechanism of increased salivation evoked by vibrotactile stimuli,
we recorded changes in brain blood flow (BBF) at the frontal cortex and the pulse frequency
during stimulation. When subjects listen to classical music (particularly Mozart), they develop
a relaxed feeling. Specifically, the feeling of relaxation is produced by decreasing BBF in the
frontal cortex. In particular, we suggest that the relaxed feeling is produced by an increase in
parasympathetic activity. Furthermore, we examined changes in the pulse frequency during
vibrotactile stimulation. A decrease in pulse frequency suggests an increase in
parasympathetic activity (Principles of Neural Science. 2000b). Thus, we assumed a
mechanism of increased salivation by exploring oxyhemoglobin (oxyHb) concentration in the
BBF of the frontal cortex and changes in pulse frequency. We believe that the coordination is
carried out by a highly interconnected set of structures in the brain stem and forebrain that
form a central autonomic network (Principles of Neural Science. 2000b).
2. Material and methods
2.1 Vibrotactile stimulation apparatus
The vibrotactile stimulation apparatus consists of an oscillating body and control unit, as
shown in Hiraba et al. (2008) and Yamaoka et al. (2007). The oscillating body is composed of
the headphone headset equipped with vibrators as a substitute for positions of the bilateral
microphones, and vibrators utilizing the vibration electric motor (VEM) (Rekishin Japan Co.,
LE12AOG). The VEM was covered in silicon rubber (polyethyl methacrylate, dental mucosa
protective material, Shyofu Co.) for conglobating the stimulation parts and preventing the
warming of the VEM's temperature produced by the vibration of long periods (Hiraba et al.
2008). The control unit consists of three parts, the pulse width modulation (PWM) circuit in
Figure 1A-a, LCD monitor circuit (Figure 1A-b) and power supply circuit (Figure 1A-c), and
it interfaced with a PWN electric motor, delivered vibration frequencies in the 60-182 Hz
range (Yamaoka et al. 2007).
We examined the amount of salivation during vibrotactile stimuli on the bilateral masseter
muscle belly (on the parotid glands) and on bilateral parts of the submandibular angle (on
the submandibular glands; Fig. 1B, 1C). We determined the amount of salivation using a
dental cotton roll (1 cm across, 3 cm length) positioned at the opening of the secretory ducts
(right and left sides of the parotid glands and right and left sides of the submandibular and
sublingual glands), during vibrotactile stimulation of the bilateral parotid and
submandibular glands. The weights of the wet cotton rolls after 3 min of use were compared
with their dry weights (Hiraba et al. 2008).
2.2 Stimulating salivation in normal subjects
We determined that a 3-min salivation measurement with a 5-min recovery time was
sufficient from a previous experiment (Hiraba et al. 2008). First, we used three frequencies
with a single motor (89, 114, 180 Hz-S) on the parotid glands (Fig. 1B, 1D) and conducted a
practice exercise so that the participants could learn to avoid the foreign-body sensation of
the cotton rolls for 3 min. Next, after a 5-min rest, we examined the amount of salivation
during 89 Hz-S vibrotactile stimulation for 3 min. After every 5 min of rest, we examined the

Changes in Brain Blood Flow on Frontal Cortex Depending on Facial Vibrotactile Stimuli 339
amount of salivation during 114 and 180 Hz-S vibrotactile stimulation for 3 min,
respectively. We finally examined the amount of salivation on resting stimulation for 3 min
(Fig. 1D). We conducted these tests with 19 normal subjects (6 males, 13 females; average
age, 22 years). This experiment was performed between 3 and 5 pm in a temperature-
controlled, quiet room.

Fig. 1. Pictures and diagrams of the vibratory machine (A), experimental techniques (B and
C) and experimental schedules (D and E). B and C. Measuring salivation amount during
vibrotactile stimulation of the bilateral masseter muscle belly (on the parotid glands, B) and
the bilateral angle of the mandibular body (on the submandibular glands, C), using dental
cotton rolls positioned at the opening of the secretory ducts (right and left sides of parotid
glands and right and left sides of submandibular and sublingual glands). D and E. Time
schedules for the vibrotactile stimuli, respectively. The designations 89 Hz-S, 114 Hz-S and
180 Hz-S indicate vibrotactile stimuli at 89, 114, and 180 Hz, respectively, with a single
motor. The designations 89 Hz-D and 114 Hz-D represent vibrotactile stimuli of 89 and 114
Hz with double motors. We allowed a 5-min recovery time shown as rest hours. Variation
per day in the effective salivation during 4 or 5 days of continuous use at 89 Hz-S
vibrotactile stimulation for the right and left parotid glands, and the time schedule as shown
in Figure 1D was repeated by subjects for 4 or 5 days.
After the three frequencies were tested on the parotid glands, we determined the most
effective frequencies, which were 89 and 114 Hz-S. Second, we subjected the submandibular
glands to the two frequencies (89 and 114 Hz) and two different amplitudes (Fig. 1C, 1E).

The amplitude of the oscillating bodies during vibrotactile stimulation was measured with a
CCD laser displacement gauge (LK-G3000, Keyence Co., Osaka, Japan). We examined the 89
and 114 Hz frequencies and used oscillating bodies added as the frequency with double
motors (single motor had a 1.9 m amplitude, 89 and 114 Hz-S, and double motors had a 3.5
m amplitude, 89 and 114 Hz-D). Specifically, 89 and 114 Hz-S, and 89 and 114 Hz-D were
used in the next experiment. We examined the amount of salivation for the four frequencies
(Fig. 1E). We conducted these experiments with 17 normal subjects (15 males, 2 females;
average age, 22 years) for the resting-stimulation examination. This experiment was
performed between 3 and 5 pm in a temperature-controlled, quiet room.
The most effective salivation amount by vibrotactile stimulating the parotid glands occurred
at 89 Hz-S, so we examined changes in salivation for 4 or 5 continuous days with the same
time schedule (Fig. 1D). We also investigated the adaptation periods during continuous use
of vibrotactile stimulation for 4 or 5 continuous days in the same subjects. We conducted
resting-stimulation examinations with 26 normal subjects (11 males and 15 females; average
age 25 years). This experiment was performed between 3 and 5 pm in a temperature-
controlled and quiet room.
2.3 BBF on the frontal cortex
The recording was conducted using a functional near-infrared spectroscopy (fNIRS) OEG16
instrument (Spectratech Inc., Shelton, CT, USA) from the frontal cortex. As shown in Figures
4A and 4B, the fNIRS probe assembly consisted of six LEDs as light sources, each of which
emitted two wavelengths, 770 nm and 840 nm, and six photodiodes as detectors (Fig. 4A,
4B). The sources and detectors were symmetrically arranged in an area of 3.014.0 cm, with
a nearest source-detector separation of 2.0 cm, and measurement points were at 16 points on
the frontal cortex. During scanning, a Velcro band held the probe assembly securely to the
forehead of subjects and extended from ear to ear horizontally and from hairline to
eyebrows vertically. Each of the LEDs was turned on in sequence, and the diffuse NIR light
from each source was acquired through the cortical region at the nearest detector. Thus, 16
source-detector pairs (channels) in total were measured (Fig. 4B). The sampling rate across
all 16 channels was 0.76 Hz. In particular, we showed a 16-channel computerized analysis
(as shown in Fig. 4B) and the original waves of four channel recording areas in the central
parts (original waves in Fig. 4C recorded over the selected areas in Fig. 4A).
We recorded BBF during vibrotactile stimulation. We also conducted a BBF experimental
recording as subjects listened to classical music (Mozart, Eine kleine Nachtmusik). We
conducted the examinations with 10 normal subjects (6 males, 4 females; average age, 22
years) for the resting-stimulation and classical music examinations. This experiment was
performed between 3 and 5 pm in a temperature-controlled, quiet room.
2.4 Pulse frequency during vibrotactile stimulation
We recorded changes in pulse frequency (P225F, Nihon-Kohden Co., Tokyo, Japan) during
vibrotactile stimuli at 89 and 114 Hz-S. Because the hemoglobin in blood absorbs red light
irradiation, a pulse wave was calculated using a ratio between red light irradiation (660 m)
and infrared light (940 m). The pulse frequency indicated by this apparatus was measured
by averaging the pulse frequency during the previous eight pulses (Fig. 5C). First, the pulse
frequency during rest was recorded for 1 min, and then those during 89 and 114 Hz-S
vibrotactile stimuli were recorded for 2 min each. A 1-min rest period was taken between

the 89 and 114 Hz-S vibrotactile stimuli. Finally, the pulse frequency during rest was
recorded for 30 s. We conducted these examinations with 10 normal subjects (6 males, 4
females; average age, 22 years). This experiment was performed every day between 3 and 5
pm in a temperature-controlled, quiet room.
3. Results
3.1 Vibrotactile stimuli on the parotid and submandibular glands
We examined the differences between vibrotactile stimulating of parotid and
submandibular glands (Fig. 1B, 1C). First, we tested three frequencies of vibrotactile
stimulation, 89, 114, and 180 Hz-S, on the parotid glands and determined the frequency
most effective for salivation (Fig. 2A). We found that the most effective frequencies for
submandibular gland salivation were 89 Hz-S, 89 Hz-D, 114 Hz-S, and 114 Hz-D (Fig. 2B)
and that the erase most effective stimulation was 89 Hz-S (Fig. 2A, 2B; p < 0.05).
We examined the most effective salivation in each gland, including the right parotid, left
parotid, right submandibular and sublingual and left submandibular and sublingual glands.
As a result, the most effective stimulation for salivation of the parotid and submandibular
glands was 89 Hz-S (Fig. 2A-a, 2B-a; resting salivation, 0.890.61 mL; 89 Hz-S, 1.460.90 mL;
89 Hz-D, 1.040.78 mL; 114 Hz-S, 0.930.76 mL and 114 Hz-D, 0.800.53 mL). Furthermore, a
significant difference was observed between the vibrotactile stimulation at 89 Hz-S and that
at 114 Hz-D (paired t-test; p < 0.05). No other significant difference in each gland was
observed (Fig. 2A-b, 2A-c, 2B-b, 2B-c, 2B-d and 2B-e).

Fig. 2. Salivation from the parotid glands (A) and the submandibular glands (B) after each
vibrotactile stimulation. A-a and B-a. Total number of salivations after each vibrotactile
stimulation. A-b, A-c, B-b, B-c, B-d and B-e. Salivation amount for each gland after various
vibrotactile stimuli. The increased salivation between 89 Hz-S and 114 Hz-D was
significantly different (paired-t-test; p < 0.05). Stimulating the parotid or submandibular
glands at 89 Hz-S was the most effective for salivation. Salivation was not different among
the various glands (parotid and submandibular and sublingual glands on the right and left
sides).
3.2 Variation in effective salivation per day during continuous vibrotactile stimulation
Because patients with hyposalivation often have a psychiatric disorder, we conducted an
experiment to realistically approximate natural conditions. We examined whether effective
salivation occurred continuously when vibrotactile stimulation was performed daily. In

particular, we used the 89 Hz-S from the previous experiment, and because no difference
was found between vibrotactile stimulation of the parotid and submandibular glands, we
analyzed salivation evoked by vibrotactile stimulation of the parotid glands. Normal
subjects (15 males, 11 females; average age, 25 years) used this apparatus continuously for 4
or 5 days at the same time and place (Fig. 3). None of the glands (right and left parotid
glands, and right submandibular and sublingual glands) showed an attenuated response.
Regression curves for each gland (Fig. 3A, 3B, 3C and 3D) showed non-adaptation to
continuous stimulation, because they showed parallel or over-increasing curves, indicating
that continuous usage of this apparatus should not be a problem.

Fig. 3. Variation in effective salivation per day during 4 or 5 days of continuous use of 89
Hz-S vibrotactile stimulation in the right (A) and left parotid (B) and the right (C) and left
submandibular and sublingual glands (D). The time schedule, as shown in Figure 1D, was
repeated by subjects for 4 or 5 days. Vertical lines indicate the increase and decrease in
vibrotactile salivation versus resting salivation (it is shown by a positive value when the
vibrotactile salivation was greater than the resting salivation, and by a negative value when
the vibrotactile salivation was lower than the resting salivation). Abscissa lines indicate time
(45 days). Solid lines indicate the regression curve. There was no adaptation to the 89 Hz-S
vibrotactile stimulation in the parotid, submandibular, or sublingual glands.
3.3 Relationship between each stimulation and fNIRS activity
The OEG16 spectroscope was used to record the hemoglobin concentration in the BBF from
areas in the frontal cortex using 16 channels. In particular, we measured original waves in
four channel recording areas (circled 1, 2, 3, and 4 in Fig. 4A). Figure 4B shows the schema of
the oxyhemoglobin (oxyHb) concentration evoked by analyzing 16 channels during
vibrotactile stimulation at 89 Hz-S. The contrasting red and blue zones indicate activity
strength, and no response is shown in white (Fig. 4B). The results showed very weak oxyHb
concentrations during the 89 Hz-S vibrotactile stimulation, as shown in Figure 4C-b.
We provided evidence of the real waves on the central parts (circled 1, 2, 3, 4; Fig. 4A) of the
BBF recording in 16 channels (Fig. 4C). The red lines indicate changes in oxyHb
concentrations, the blue lines indicate deoxyhemoglobin (deoxyHb), and the green lines
indicate total hemoglobin (totalHb) during salivation measurements at rest and for each
vibrotactile stimulation frequency. The first waves in Figure 4C(a) indicate changes during
the resting condition, the second waves (b) show changes during 89 Hz-S vibrotactile
stimulation, the third waves (c) show changes during 89 Hz-D, the fourth waves (d) show
changes during 114 Hz-S, the fifth waves (e) show changes during 114 Hz-D and the bottom
waves (f) show changes during A- phonation. Each wave was recorded for 3 min, and
each vibrotactile stimulation provided for 2 min is shown between the vertical lines.
However, A- phonation is shown by four applicable vertical lines. Although each wave
A
B
C
days
ml
1.5
-0.5
0.5
-1.5
ml
1.0
0.5
-0.5
-1.0


Fig. 4. A. Frontal cortex recording sites. The circled numbers 1, 2, 3, and 4 indicate the
recording sites for each wave, as shown in C. B. A computerized numerical analysis of
variations in oxyhemoglobin (oxyHb) concentration during 89 Hz-S vibrotactile stimulation.
Red (plus) and blue shading (minus) indicates differences in frontal cortex activation,
whereas the white band indicates non-activation. C. Changes in blood brain flow during 2
min of various vibrotactile stimuli and vocalizing A- sounds are shown at the central
areas of the frontal cortex. C-a. During the resting condition. C-b. During 89 Hz-S
vibrotactile stimulation. C-c. During 89 Hz-D vibrotactile stimulation. C-d. During 114 Hz-S
vibrotactile stimulation. C-e. During 89 Hz-D vibrotactile stimulation. C-f. While vocalizing
A- sounds. The spaces between the solid bars (2 min.) represents the stimulating period.
However, vertical lines during vocalizing A- sounds were the timely accomplishment
between good winds. Red, blue, and green lines indicate oxyhemoglobin, deoxyhemoglobin,
and total hemoglobin concentrations, respectively.

measured during resting salivation, at 114 Hz-D, and during A- phonation showed
increased activity, the 89 Hz-D and 114 Hz-S vibrotactile stimuli showed decreased
activities. However, vibrotactile stimulation at 89 Hz-S was fairly close to zero value
(0.141.10 mM mm s). In particular, when we focused on changes in oxyHb based on these
results, increased oxyHb occurred during A- phonation, the resting condition (5.206.78
mM mm s) and at 114 Hz-D of vibrotactile stimulation (-0.133.54 mM mm s), whereas a
decrease in oxyHb was revealed during vibrotactile stimulation at 114 Hz-S (-5.487.12 mM
mm s) and at 89 Hz-D (-2.1810.13 mM mm s). However, oxyHb during vibrotactile
stimulation at 89 Hz-D (0.292.26 mM mm s) was near zero, similar to all of the data
(oxyHb, deoxyHb, totalHb). From these results, we computed integral rates of oxyHb
during 2 min, as shown by the area between the longitudinal bars in the central part (circled
1, 2, 3, and 4) of the recording wave (Fig. 4C).
Furthermore, we examined the integral rates while subjects listened to classical music for 2
min. We divided the subjects into two groups: one group did not like to listen to classical
music (2.966.40 mM mm s, n = 7) whereas the other group did (0.850.05 mM mm s, n = 3;
Fig. 5A and 5B, respectively). Although the subjects who enjoyed the music did not show a
larger spread of values (SD), the latter did. In particular, vibrotactile stimulation at 89 Hz-S
revealed a small similar spread of values. Each integral rate during the vibrotactile
stimulation at 89 Hz-S and while the listening to classical music showed similar averages
and SDs (Fig. 5B-b).
3.4 Pulse frequency during vibrotactile stimulation and frontal cortex fNRS activity
We chose three typically responding subjects, as shown in Figure 5C. One subject showed a
decrease in pulse frequency during both 89 Hz-S and 114 Hz-S vibrotactile stimulation,
while the other two showed a decrease in pulse frequency at 89 Hz-S and an increase during
114 Hz-S stimulation (Fig. 5C). We divided the two groups into decrease pulse frequency
and increase pulse frequency groups, based on 114 Hz-S vibrotactile stimulation. Although
the 89 Hz-S vibrotactile stimulation created a sense of relaxation, the 114 Hz-S vibrotactile
stimulation is either liked or disliked. Specifically, the 89 Hz-S vibrotactile stimulation will
produce parasympathetic activity and evoke a relaxation response.
4. Discussion
4.1 Adaptation to vibrotactile stimulation
Continuous exposure to various sensory stimuli induces adaptation (Principles of Neural
Science. 2000c). Thus, we examined whether adaptation occurred to the vibrotactile
stimulation when normal subjects continuously used this apparatus every day for 4 or 5
days. However, we first investigated changes in total saliva secretion quantities by normal
subjects in response vibrotactile stimulation by measuring the weights of cotton rolls set on
the ducts of each gland. For the repeated measurements, we defined a recovery time of 5
min from a previous experiment, as resting salivation values before and after stimulating
salivation were the same (Fig. 1D, 1E).
The most effective frequency for increasing the total amount of salivation during vibrotactile
stimulation of the parotid and submandibular glands was 89 Hz-S (Fig. 2A-a, 2B-a).
Furthermore, as shown in Figures 2A-b and 2A-c, and 2B-b, 2B-c, 2B-d and 2B-e, no
difference was found among the various glands. Because 89 Hz-S vibrotactile stimulation of
the parotid and submandibular glands showed the same results in all glands, the results

might not be directly affected by the vibrotactile stimulation on the parotid and
submandibular glands. Specifically, the increased salivation evoked by vibrotactile
stimulation at 89 Hz-S may be the result of somatosensory input from the facial skin and
intraoral cavity.
Burdette and Gale (1988) reported that tonic masticatory muscle activity may be effective for
treating patients with myofacial pain dysfunction. Furthermore, Vrjama and Vanharantra
(1994) reported that discographically painful discs always decrease pain during a vibration
examination. These results assume that the peripheral stimuli provided by the vibration
arrive at the central nervous system (spinal cord and brain stem) and that these effects are
produced by a depressant effect on the cerebral cortex, depending on the somatosensory
information, although they are not clearly organized. We believe that their interpretation
does not adequately explain this situation. We suggest that vibrating stimuli may promote
parasympathetic effects by inhibiting the sympathetic effects elicited by pain. We know
serous salivation is evoked only by a parasympathetic effect. Furthermore, we found that
salivation production responded to a specific frequency and amplitude.
Vibration may effectively activate the muscle spindles of the masseter muscles, because
the tonic vibration reflex (TVR) of the muscle spindles activates with 80100Hz vibrations
(Desmedt, Codaux 1975). However, we showed the same salivation response in various
glands. We showed that increased salivation evoked by 89 Hz-S vibrotactile stimulation
would be produced by somatosensory inputs from facial skin and the intraoral cavity,
suggesting that increased salivation is not produced by a direct effect of muscle
contraction.
Next, we examined the physiological characteristics of adaptation to vibrotactile stimulation
by testing adaptation in 26 normal subjects for 4 or 5 days. Patients with decreased
salivation (hyposalivation) would be using this machine daily; thus, we wanted to
realistically approximate actual conditions. The results suggested that 89 Hz-S vibrotactile
stimulation to the facial skin of the masseter belly may be appropriate for patients with
decreased salivation, because the regression curve (negative gradient) did not decrease with
continuous use of the apparatus (Fig. 3).
4.2 fNIRS activity in the frontal cortex
The frontal cortex is associated with cognitive function, including memory, attention,
abstract reasoning and higher cognitive activity (Principles of Neural Science. 2000a).
We recorded changes in BBF in the frontal cortex to examine typical changes in fNIRS
parameters based on increases in oxyHb and totalHb and decreased deoxyHb, as reported
by Sakatani et al. (2006). In particular, increased fNIRS activity patterns in the frontal cortex
are associated with speech, and decreased fNIRS activity patterns in the frontal cortex are
associated with playing TV games and are thought to be dependent on network differences
(Sakatani et al. 1999). The fNIRS responses may be influenced by mental stability, because
they indicate control of sophisticated mental functions produced by complex networks.
Because fNIRS activity increases while subjects are speaking and decreases while they play
TV games, we suggest that the autonomic system has a profound effect. Specifically, we
suggest that fNIRS activity during conscious speaking is associated with a sympathetic
effect, and that playing a TV game is associated with reflective movement under a non-
sympathetic effect.
The effect of 89 Hz-S vibrotactile stimulation was almost zero for oxyHb, deoxyHb, and
totalHb (Fig. 4C-b). In particular, the fNIRS activity focuses on the excitatory behaviours to

increase oxyHb. In animal experiments, changes in oxyHb and BBF are related, and fNIRS
activity changes in oxyHb are used as a neuronal activity index (Hoshi et al. 2001). Thus,
changes in oxyHb produced by an 89 Hz-S vibrotactile stimulation on the parotid and
submandibular glands may indicate mental stability. The reason may be due to the same
tendency of oxyHb concentration between the 89 Hz-S vibrotactile stimulation and subjects
who liked to listen to classical music (Fig. 5B-b). People relax when they listen to classical
music, so we think that 89 Hz-S vibrotactile stimulation is produced by excitation of the
parasympathetic system. In particular, although the 89 Hz-S vibrotactile stimulation always

Fig. 5. Integral values of oxyhemoglobin (oxyHb) concentration produced by vibrotactile
stimuli for 2 min are shown in A and B. A. Integral values produced in the resting condition
and after each stimulation (n = 17). B. Integral values produced in the resting condition,
listening to classical music (Mozart, Eine kleine Nachtmusik), 89 Hz-S and listening to noise
(n = 10). One group enjoyed listening to classical music (D-b, n = 3), and one did not (D-a, n
= 7). C. Changes in pulse frequency during vibrotactile stimulation (89 Hz-S and 114 Hz-S).
We assessed 10 normal subjects, and changes in pulse frequencies are shown for three
typical subjects. All subjects had decreased pulse frequency during 89 Hz-S vibrotactile
stimulation, although the 114 Hz-S vibrotactile stimulation divided the subjects into two
groups: one showed an increased pulse frequency and the other showed a decreased pulse
frequency.

revealed parasympathetic excitation, listening to classical music showed different activity
depending on music preference (Fig. 5B-a, 5B-b). Those subjects who enjoyed Mozart
classical music accepted it as relaxing, whereas those who disliked it perceived it as noise.
However, 89 Hz-S vibrotactile stimulation may produce a balanced mental condition,
regardless of preference. This phenomenon suggests that the effect caused by the 89 Hz-S
vibrotactile stimulation and the feeling sensed by those listening to Mozart who enjoyed it
may be the same. Thus, we suggest that these feelings were produced by parasympathetic
activity.
Burdette and Gale (1988) reported that tonic masticatory muscle activity might be effective
for treating patients with myofascial pain-dysfunction. Furthermore, Vrjama and
Vanharantra (1884) reported that discography always results in less pain during a vibration
examination. They suggested that peripheral stimuli produced by the vibration arrive at the
central nervous system (spinal cord and brain stem) and that these effects are caused by a
depressant effect on the cerebral cortex, depending on the somatosensory information
(Burdette and Gale, 1988; Vrjama and Vanharantra, 1884). However, we believe that
somatosensory information via the thalamus produces excitability in the primary
somatosensory cortex. Furthermore, somatosensory information in the primary
somatosensory cortex may project to the frontal cortex via the parietal association area
(Handbook of Neuropsychology 1994), and this information may cause the excitatory
effects.
4.3 Parasympathetic effect based on 89 Hz-S vibrotactile stimulation
When we are frightened, our heartbeat increases (Principles of Neural Science. 2000b). The
parasympathetic nervous system is responsible for rest and digestion, and maintaining basal
heart rate, respiration, and metabolism under normal, resting conditions (Principles of
Neural Science. 2000b). We examined the parasympathetic effects with changes in pulse
frequency during vibrotactile stimulation. We showed changes in pulse frequency in typical
normal subjects in response to 89 Hz-S and 114 Hz-S vibrotactile stimuli (Fig. 5C). These
results showed a decrease in pulse frequency during 89 Hz-S stimulation, but an increase in
pulse frequency following the 114 Hz-S vibrotactile stimulation, except for one subject.
Specifically, the 114 Hz-S vibrotactile stimulation showed variable results, because one
subject was relaxed, whereas the others were not. However, the 89 Hz-S vibrotactile
stimulation resulted in decreased pulse frequency in all subjects, suggesting that stimulating
the face evoked parasympathetic stimulation (Fig. 5C). Thus, the parasympathetic activity
elicited by 89 Hz-S vibrotactile stimulation, may produce an increase in salivation. We
reiterate that a relaxed feeling was produced in all subjects during 89 Hz-S vibrotactile
stimulation.
We consider the following schema for this phenomenon. Vibrotactile stimulation to the face
produces somatosensory information activity in the orofacial region, which projects to the
trigeminal sensory complex (principal trigeminal sensory nuclei and trigeminal spinal tract
nuclei) and solitary nuclei (Fig. 6). This somatosensory information is divided into the
parabrachial nuclei, thalamus, and hypothalamus. Information projecting to the
parabrachial nuclei arrives in the hypothalamus, amygdala, and frontal cortex and then to
the thalamus and primary somatosensory cortex (SI). Petrides reported that information
travelling to the SI goes around the frontal cortex via the parietal association area
(Handbook of Neuropsychology 1994). We think that somatosensory information travelling

via the parietal association area represents excitability. If the BBF in the frontal cortex via the
parietal association area is excitable and that via the parabrachial nucleus is inhibitory,
which wave forms will be revealed? We believe that BBF to the frontal cortex evoked by
various stimuli shows various wave forms, depending on the condition. In particular, the 89
Hz-S vibrotactile stimulation may produce a relaxed feeling.
4.4 Projection of autonomic activity in the frontal cortex
Autonomic function must ultimately be coordinated by adapting to environmental
changes. The autonomic nervous system is a visceral sensory and motor system, and the
visceral reflex is mediated by local circuits in the brain stem or spinal cord. These reflexes
are regulated by networks of central autonomic control nuclei in the brain stem,
hypothalamus and forebrain and are not under voluntary control, nor do they impinge on
consciousness, with a few exceptions (Principles of Neural Science. 2000b). However, we
think that changes in BBF in the frontal cortex may represent autonomic activity. This
coordination is carried out by a highly interconnected set of structures in the brain stem
and forebrain that form a central autonomic network (Fig. 6 and Principles of Neural
Science. 2000b). Furthermore, the key component of this network is initiated by integral
information in the parabrachial nucleus of the solitary tract and trigeminal sensory
complex in the brain stem. These nuclei receive inputs from somatosensory and visceral
afferents of the trigeminal, facial, glossopharyngeal, and vagus nerves and then use the
information to modulate autonomic function. The somatosensory and visceral sensory
outputs from the trigeminal and solitary nuclei are relayed to the forebrain and amygdala
by the parabrachial nucleus, which is important for the behavioural responses to
somatosensory, taste, and other visceral sensations (Principles of Neural Science. 2000b),
and the information arriving in the amygdala will produce the pleasure-pain feeling. In
contrast, the parabrachial nucleus is a taste-sensation relay nucleus in rats (Scott and
Small 2009), and the rodent parabrachial nucleus sends integral limbic and reward system
information (Yamamoto et al. 2009). Although their function in humans is unknown, we
think that these nuclei may play a role as a relay nucleus for the autonomic system. On
the other hand, we showed a projection from the trigeminal sensory complex, as the
parabrachial nucleus can also record the response to tactile stimuli from facial skin
(Chiang et al, 1994). Furthermore, somatosensory information is projected to the primary
somatosensory cortex and is then relayed to the frontal cortex, via the parietal association
area (Handbook of Neuropsychology 1994).
What does BBF activity in the frontal cortex mean? We think that the information via the
parabrachial nucleus dominates that via the parietal association area. Thus, information in
the frontal cortex is assumed to arrive via the parabrachial nucleus area. The hypothalamus
is the centre of the autonomic system. We perceive emotional experiences such as fear,
pleasure, and contentment and these reflect the interplay between higher brain centres and
sub-cortical regions, such as the hypothalamus and amygdala (Principles of Neural Science.
2000a). Patients in whom the prefrontal cortex or the cingulate gyrus has been removed are
no longer bothered by pain, but exhibit appropriate autonomic reactions; however, the
sensation is not perceived as a powerfully unpleasant experience (Principles of Neural
Science. 2000a). Furthermore, the anatomical connections of the amygdala with the temporal
(cingulate gyrus) and frontal (prefrontal) association cortices provide the means by which
visceral and somatosensory sensations trigger a rich assortment of associations or the
cognitive interpretation of emotional states (Principles of Neural Science. 2000b).


Fig. 6. Pathways distributing somatosensory information in the brain. Somatosensory
information evoked by vibrotactile stimulation is relayed by the trigeminal sensory complex
and solitary and parabrachial nuclei, which arrives at the hypothalamus, thalamus,
amygdala and frontal cortex, respectively. The autonomic system (particularly the
parasympathetic nervous system) produces increased salivation. The lateral branch of the
trigeminal sensory nucleus is projected to the parabrachial nucleus. Information from the
parabrachial nucleus is received by the amygdala and frontal cortex. Furthermore,
somatosensory information is projected to the primary somatosensory cortex and relayed to
the frontal cortex via the parietal association area. Thus, this information finally leads to a
relaxed feeling and BBF waves reflect parasympathetic activity (modified from schemas in
Principles of Neural Science 2000b and Handbook of Neuropsychology 1994).

Finally, we showed oxyHb activity during A- phonation (Fig. 4C-f). Sympathetic activity
increases when we speak. From our data, the facial stimulation effects evoked by 89 Hz-S
vibrotactile stimulation evoked parasympathetic activity, depending on the decrease in
pulse frequency, and oxyHb concentration reflected the non-activation of BBF waves in the
frontal cortex. We suggest that the 89 Hz-S vibrotactile stimulation produced a relaxing
sensation depending on excitation in the amygdaloid complex and parasympathetic activity
in the hypothalamic nuclei. The relaxing sensation was reflected by the oxyHb concentration
of the BBF wave in the frontal cortex.
5. Conclusion
We showed that the most effective salivation response was produced by an 89 Hz
vibrotactile stimulation with a single motor (1.9 m amplitude) on the parotid glands, as
reported previously (Hiraba et al. 2008). In this study, we focused on the submandibular
glands secreting the greatest amount of saliva and explored changes in salivation during
stimulation of the submandibular glands with 89 Hz and 114 Hz vibrotactile stimuli using
single (89 Hz-S and 114 Hz-S) or double motors (3.5 m amplitude, 89 Hz-D and 114 Hz-D),
respectively. The increased salivation evoked by the 89 Hz-S vibrotactile stimulation was
produced by somatosensory inputs from the facial skin and intraoral cavity. We examined
the effects of vibrotactile stimulation on adaptation in normal subjects to test the effect of
continuously using this apparatus every day by patients with hyposalivation. The results
showed no adaptation with continuous use. Furthermore, we investigated changes in BBF in
the frontal cortex during vibrotactile stimulation. Stimulating the submandibular and
parotid glands at 89 Hz-S resulted in the most salivation. OxyHb concentrations in the BBF
to the frontal cortex during an 89 Hz-S vibrotactile stimulation were the same as those in
subjects who preferred listening to classical music. Furthermore, parasympathetic activity
decreased pulse frequency during an 89 Hz-S vibrotactile stimulation, suggesting a close
relationship between the BBF waves in the frontal cortex and the autonomic system. An
89 Hz-S vibrotactile stimulation may produce relaxation and salivation may be increased by
parasympathetic excitation.
6. Acknowledgements
This work was supported by a Sogoshigaku research grant and the Sato Fund of Nihon
University School of Dentistry, as well grants from the Ministry of Education and Grants-in-
Aid for Scientific Research (21592539).
7. References
Ueda K. 2005. Sessyoku enge rehabilitation. In: Uematsu H, Inaba S, Watanabe M, editors.
Koureishya Shika guidebook. Tokyo: Ishiyaku. pp 248275 (in Japanese).
Hiraba H., M. Yamaoka, M. Fukano, K. Ueda and T. Fujiwara (2008) Increased secretion of
salivary glands produced by facial vibrotactile stimulation. Somatosensory and
Motor Research 25, 222229.

Despopoulos, A. and Silbernagl, S (2003) 10 Nutrition and digestion. In: Color Atlas of
Physiology. 5
th
ed. New York, Medical science international Ltd. pp
226265.
Yamaoka, M., Hiraba, H., Ueda, K. and Fujiwara, T. (2007) Development of a vibrotactile
stimulation apparatus for orofacial rehabilitation. Nihondaigaku Shigakubu Kiyou 35 ,
pp. 1318. (in Japanese)
Burdette BH, Gale EN. (1988) The effects of treatment on masticatory muscle activity and
mandibular posture in myofascial pain-dysfunction patients. J Dent Res 67; 1126
1130.
Vriama M, Vanharanta H.(1994) Bony vibrotactile stimulation: A new, non-invasive method
for examining intradiscal pain. Eur Spine J 3; 233235.
Desmedt JE, Codaux E. (1975) Vibration induced discharge patterns of single motor unit in
the masseter muscle in man. J Physiol 253; 420442.
Ivarsen S., Kupfermann F., Kandel ER. (2000a) 50, Emotional States and Feelings. In: Kandel
ER., Schwartz JH., Jessell TM., editors, Principles of neuronal science, 4
th
ed, New
York: McGraw-Hill. pp 982997.
Sakatani K, Yamashita D, Yamanaka T (2006) Changes of cerebral blood oxygenation and
optical path length during activation and deactivation in the prefrontal cortex
measured by time-resolved near infrared spectroscopy. Life Sciences 78;
27342741.
Sakatani K, Lichty W, Xie Y (1999) Effects of aging on language-activated cerebral blood
oxygenation changes of the left prefrontal cortex. Near infrared spectroscopy
study. J Stroke Cerebrovascular Dis. 8; 398403.
Hoshi Y, Kobayashi N, Tamura M. (2001) Interpretation of near-infrared spectroscopy
signals: A study with a newly developed perfused rat brain model. J Appl Physiol.
90; 16571662.
Ivarsen S., Ivarsen L., Saper CB. (2000b) 49, The Autonomic Nervous System and the
Hypothalamus. In: Kandel ER., Schwartz JH., Jessell TM., editors, Principles of
neuronal science, 4
th
ed, New York: McGraw-Hill. pp 960981.
Gardner EP, Martin JH. (2000c) 21, Coding of sensory information. In: Kandel ER., Schwartz
JH., Jessell TM., editors, Principles of neuronal science, 4
th
ed, New York: McGraw-
Hill. pp 411429.
Chiang CY, Hu W and Sessle BJ (1994) Parabrachial area and nucleus raphe magnus-
induced modulation of nociceptive and non-nociceptive trigeminal subnucleus
caudalis neurons activated by cutaneous or deep inputs. J. Neurophysiol. 71,
24302445.
Petrides M (1994) Frontal lobes and working memory: evidence from investigations of the
effects of cortical excisions in nonhuman primates. In: Handbook of
Neuropsychology, Vol. 9, F. Boller and J. Grafman (eds.), Elsevier Science B.V.,
Amsterdam, pp. 5082.
Yamamoto T, Takemura M, Inui T, Torii K, Maeda N, Ohmoto M, Matumoto I, Abe K. (2009)
Functional organization of the rodent parabrachial nucleus. Ann N Y Acad Sci.
1170; 378382.

Scott TR and Small DM (2009) The role of the parabrachial nucleus in taste processing and
feeding. Ann N.Y. Acad Sci. 1170, 372377.

Vibration Analysis and Control PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Vibration Analysis and Control PDF

Uploaded by

Copyright:

Available Formats

VIBRATION ANALYSIS AND

CONTROL NEW TRENDS

using the Real Time Workshop

and Real Time Windows Target

, the tuned acceleration at A at an excitation

denotes the torsional

denotes the summation of damping of platform and additional

expresses the stiffness of platform; u

is the control force of CTLCD to offshore

= denotes inertia moment ratio. Let ( )

minimum, so the optimal parameters of CTLCD can be obtained according to

and solving above equation, the

, the optimal damping ratio of CTLCD

, the equivalent damping ratio

are transfer functions from

= is the three-dimensional siesmic inputs;

= " " " is the three-dimensional control vector, where

are the drifts of a story in x, y and torsion direction.

G j (Eq. (5)) is obtained with

14.74 rad s = . The compensator parameters are thus

are the vertical displacements of sprung mass,

(i = 1, 2, S, 4) are sliding surface coefficients to be determined so that the sliding

negative, the following control law is proposed to generate the

control of active vehicle suspensions with actuator time

Fig. 4. Nominal and optimal stiffness parameters k

is a constant controlling the slope of , where E is the elastic modulus of

T loads on the blade.

control on a WT model (see Connor et al., 1992). Their

control can be used for load reduction has later

phase margin in each input channel (Saberi

phase changes can be tolerated in each input

-optimal control approach

controller design, z is called the output performance vector to

x = Ax +Bu + H(y Cx) (7)

x = Fx +Gu + Hy, (14)

has full column rank for all

has full row rank for all

You might also like